[BidClub_]
20VC · · 101 min

Cursor Acquired for $60BN | Anthropic Hits $1TRN in Secondary Markets & Figma, Adobe, Canva Dead?

Harry StebbingsJason

YouTube ↗
TL;DR

  • The proposed $60 billion Cursor/xAI-SpaceX tie-up is an IPO-contingent option, not a transaction closing today. SpaceX can acquire Cursor after going public or pay a $10 billion break clause; Jason Lemkin sees meaningful execution conditions, but Rory O’Driscoll thinks the public promotion signals intent to close. Industrially, Cursor brings several billion dollars of coding revenue but weak gross margins, while xAI brings Colossus, “hundreds of thousands of GPUs,” roughly $20 billion of infrastructure and little revenue — “a marriage made in heaven.”

  • SpaceX’s extraordinary multiple makes Cursor cheap even if $60 billion sounds unprecedented. At an alleged $2 trillion valuation, SpaceX would spend about 3% of its equity value to acquire perhaps 15%-20% of combined revenue; even at $1 trillion, Jason’s comparison is stock trading near 50X revenue buying Cursor near 10X year-end revenue. “If your stock is valued at 100 times revenues, you can buy things that are trading at 10 or 15 times revenue all fucking day long.”

  • The deal split the panel on whether $60 billion is an M&A high-water mark or merely the new floor for strategic urgency. Jason predicts a $100 billion acquisition within 12 months because seven roughly $2 trillion companies can spend 5% of market cap to avoid falling behind; Rory predicts this remains private M&A’s high-water mark for a decade, given the finite buyer pool and sub-10X revenue multiples elsewhere. They agree on the mechanism: every major CEO is asking, “Who the hell can we buy to get ahead?”, and Cursor has expanded the Overton window for $10 billion-$20 billion deals.

  • Anthropic’s trillion-dollar secondary price reflects peak scarcity and FOMO, not a settled judgment about the AI winner. Harry says European families want “Anthropic and nothing else,” but Jason notes Codex usage rose 50% in one month and argues the autonomous-agent war has only started: “I can’t honestly predict 30 days out.” Rory expects Anthropic to exploit the moment with an October/Q4 IPO, potentially raising at $1 trillion while accessing the equity, convertibles and debt needed for a stated $200 billion-$300 billion capital appetite.

  • Claude Design need not replace Figma, Adobe or Canva to impair their growth. Jason’s crucial distinction is that Anthropic built an application — with users, sharing, hierarchy and saved assets — rather than another prompt or purple-gradient artifact generator. It is not yet Figma or Illustrator, but its Claude Code integration lets product and engineering teams bypass slow design handoffs; over four, six or eight quarters, “if it maims you, it maims you.”

  • Tim Cook’s orderly exit masks a broader warning for pre-AI incumbents: paid retention can coexist with disappearing usage. Cook leaves Apple at 65 after market capitalization rose from roughly $350 billion to $4 trillion, with the stock barely reacting to internal successor John Ternus. Jason’s “stealth churn” test is more forward-looking: he still pays Netflix while watching YouTube, and he said Amelia had not used OpenAI for four months even though they were still paying for it — making MAUs, WAUs and DAUs growing faster than revenue a new survival indicator.

  • Rippling’s $1 billion ARR and 78% growth demolish “SaaS is dead,” but agents will reorder winners through APIs. The company reportedly accelerated from below $500 million only 11 months earlier, which Rory calls “a win against the run of play”; payroll remains deterministic, regulated infrastructure that customers will not vibe-code. Yet Jason may switch providers after six years on Brex in one week based solely on an agent/API bake-off: “Our AI VP of Finance… don’t care what the UX is.”

  • Salesforce’s headless strategy is really a bid to become the trusted control plane for enterprise agents. Its two-decade-old API can preserve the data and workflow layer as human seats disappear, but Jason sees the larger prize as “agent fabric”: real-time governance, security, context and auditability across hundreds of agents and sub-agents. Salesforce is well placed because CIOs trust incumbents, but execution must arrive before that advantage expires: “You can’t let these crazy agents run amok.”

Digest · the substance, structured for research

1. Cursor’s $60 billion deal remains an option with real closing risk

  • Harry framed the announcement as a $60 billion acquisition by xAI-SpaceX with a $10 billion break clause, but emphasized that it closes roughly six months later, after SpaceX’s planned IPO. Rory’s cleaner description: SpaceX has an option to acquire Cursor and otherwise pays $10 billion for the work.

  • Jason’s caution was explicit: “I’m not sure they’re buying them for $60 billion.” Senior engineers had already moved to xAI, Cursor had substantial model-and-compute commitments around Colossus, and its value may depend on hitting a state-of-the-art-model milestone rather than merely owning an IDE.

  • Those conditions explain why Jason “wouldn’t be surprised if it doesn’t close.” Rory allowed for a “try before you buy” structure but argued that SpaceX would have buried a vendor arrangement; promoting it across X tells the market the intended outcome is an acquisition.

  • If completed, Rory called it the largest private venture acquisition ever: $60 billion roughly two to three years after founding, versus Wiz at $32 billion and WhatsApp at $16 billion. Jason’s reaction captured the time compression: most funds take 20 years; “this will be three and a half.”

2. Vertical integration repairs complementary weaknesses in Cursor and xAI

  • Rory reduced the industrial logic to two mismatched balance sheets. Cursor has an exploding coding business, several billion dollars of revenue and potentially $6 billion by year-end, but “shitty gross margins” because it lacks sufficient proprietary models and compute.

  • xAI has the inverse problem: roughly $20 billion invested in Colossus, “literally hundreds of thousands of GPUs,” substantial ongoing burn and relatively little revenue. It could sell white-label capacity, but Cursor places that infrastructure directly inside “the mother of all AI markets,” coding.

  • Rory’s stylized combination was a company making $3 billion while consuming $3 billion in gross-margin costs attached to another burning roughly $18 billion. Integration does not cover the xAI cost base, but it converts an awkward frontier-lab story into a revenue-bearing, full-stack AI business.

  • The IPO sequencing is central. SpaceX can tell public investors that xAI has a remediation plan without delaying the offering: close the flotation, exercise the option and present a $4 billion-$5 billion revenue AI platform whose Claude Code analogue is Cursor.

3. SpaceX’s premium stock turns an unprecedented price into cheap currency

  • Jason rejected Harry’s demand to name which side won: both did, because future SpaceX shareholders supply the arbitrage. At a $2 trillion valuation near 100X revenue, a $60 billion purchase is approximately 3% of market capitalization for perhaps 15%-20% of total revenue.

  • Even Jason’s downside case preserves the logic. If SpaceX trades at $1 trillion, roughly 50X revenue, while Cursor reaches $6 billion and sells for 10X, SpaceX is still buying lower-multiple revenue with higher-multiple paper: “It’s a great currency.”

  • The acquisition also solves Cursor’s buyer scarcity. Only a handful of companies could write a $60 billion check for a roughly break-even-gross-margin business; several would face Department of Justice constraints, while committee-led companies would struggle to approve it. Founder control makes the exceptional check possible.

  • Rory’s seller-side call was simpler: take the gift. Founder exhaustion becomes visible around years four or five, when leaders must choose another “tour of duty”; Cursor’s founders can avoid a second tour after three years with a $60 billion outcome.

4. Cursor investors may receive spectacular returns wrapped in SpaceX risk

  • Jason expects Cursor investors to receive SpaceX stock rather than cash. A proposed $75 billion IPO at $2 trillion creates a tiny float, so adding Cursor shares to a market already facing employee liquidity and lockup management could leave holders restricted for 12-18 months.

  • That exposure matters most to late investors. Jason put the Series B near 20X and the most recent round near 2X; Rory guessed the Series A could be 40X-50X. Jason’s example was that an 80X could become a 40X and still be a strong outcome, while a 2X entry could break even or go down.

  • Thrive’s construction impressed Rory more than a single headline multiple. It joined the A, led the B and kept deploying until roughly $800 million produced an aggregate 4X-5X outside the early stake — combining a small venture-style outlier with growth-stage dollars at enormous scale.

  • That illustrates why pure multiple rankings mislead institutional LPs. A small emerging manager might return 16X on a limited allocation, while Thrive can turn billions into several billions; “different products for different folks,” with the latter potentially more consequential to a large endowment.

5. Cursor widened the M&A window even as the panel split on $100 billion

  • Jason predicts a $100 billion acquisition within 12 months and called Cursor “one of three.” Harry’s premise was that Nvidia, Apple, Meta, Amazon, Alphabet, Microsoft and potentially SpaceX can spend 5% of a $2 trillion market cap to prevent strategic irrelevance.

  • Rory initially predicted Cursor would remain “the high water mark of private M&A for a decade.” Most plausible buyers trade below 10X revenue, making $100 billion hard to justify unless the target already produces $5 billion-$10 billion; a JPMorgan-Stripe combination made him acknowledge an exception rather than abandon the broader skepticism.

  • Jason’s boardroom evidence was experiential: senior executives under technological threat repeatedly ask, “Who the hell can we buy to get ahead?” Tuck-ins do not move a $2 trillion company, but a target above the strategic line can justify 5% dilution immediately.

  • Both agreed Cursor changed corporate psychology. A $10 billion-$20 billion acquisition now looks routine, and corp-dev teams do not treat equity as founders treat personal cash: “What does it take to buy the next Cursor? … Done.”

6. Tim Cook exits at a high while stealth churn threatens incumbents

  • Apple’s transition earned near-unqualified praise. Cook leaves at 65 for internal successor John Ternus, an Apple veteran of 24 years; despite unexpected timing, the stock moved less than 0.5%, which Rory read as evidence of an “organized, well-managed transition.”

  • Rory’s scorecard was formidable: revenue and operating income rose roughly 3X-4X, market capitalization climbed from about $350 billion to $4 trillion, and the market cap increased 12X. Because buybacks reduced the share count, he put the stock-price gain near 20X.

  • Jason nevertheless linked Cook, Reed Hastings and Adobe’s leadership transition to “AI terror.” Rory separated the exposures: Adobe needs an AI strategy existentially, Apple can remain the hardware platform despite weak Siri, and Netflix still faces primarily media challenges rather than immediate AI replacement.

  • Jason’s counter was “stealth churn.” He still pays Netflix while consuming YouTube, including more AI-generated content; he said Amelia had not used OpenAI in four months despite the subscription still being paid. He now wants MAUs, WAUs and DAUs growing faster than revenue, because billing can conceal behavioral abandonment.

7. Anthropic’s trillion-dollar heat does not settle the model race

  • Harry’s LP anecdote captured the scarcity: major European families wanted “one thing, Anthropic,” while secondary prices reached $1 trillion after the company rejected funding offers around $800 billion. The market is crediting Anthropic as the enterprise leader.

  • Jason refused to extrapolate. Codex usage had reportedly risen 50% in one month, OpenAI was launching autonomous agents as they recorded, and Jason had tested both APIs; nevertheless, he would not predict the next 30 or 90 days.

  • He dismissed stacks of Mac Minis and macros as “performative art” because little had shipped. The durable contest begins when autonomous agents work continuously, and both OpenAI and Anthropic are only entering that phase: “That war’s just started.”

  • Rory separated company quality, market price and realized return. Anthropic can be an extraordinary business, trade at $1 trillion and deliver a fabulously successful IPO, while buyers at that valuation still regret the position six months later.

8. Peak FOMO makes an Anthropic IPO more rational than another private round

  • Rory expects Anthropic to go public in October or Q4 “as soon as is humanly possible.” Rejecting $800 billion makes sense if public investors would currently fight for shares at $1 trillion, particularly if a private round carries rights the company no longer needs to grant.

  • Compute turns timing into strategy. Anthropic’s answer to OpenAI’s infrastructure advantage is better software plus as much capital as possible; if management believes it needs another $200 billion-$300 billion, public status expands access to follow-ons, convertibles, preferred structures and debt.

  • Jason noted the anomaly that a private round had reached $122 billion while SpaceX’s proposed largest IPO was $75 billion. Rory still favored public markets: one more giant private financing is possible, but deep, repeatable liquidity and lending capacity reside in public markets.

  • Jason invoked Figma’s collapse as a warning; Rory reframed it. Figma priced near $35, speculative buyers drove it above $100 and it later traded around $20 as enthusiasm rotated from cloud software to AI. Even if Anthropic falls after raising $50 billion, it has cash while private competitors keep losing money: “We win.”

9. Claude Design can maim design incumbents without matching their products

  • Jason’s most important observation was categorical: “Claude Design is an application.” Anthropic added sharing, users, hierarchy and saved assets, moving beyond prompts and the recognizable Claude-artifact aesthetic of generic websites and purple gradients.

  • It is not yet equivalent to Figma, much less Illustrator, and Jason rejected claims that it suddenly replaces elite human taste. The threat is that ordinary product teams can create production-ready work themselves rather than waiting weeks for a designer to return a file they must translate into code.

  • His horizon was deliberately measured: Claude Design may not affect Figma’s next quarter, yet over four, six or eight quarters it can “maim its growth.” Anthropic never needs to build a direct Figma or Adobe clone if each task completed inside Claude removes incremental demand elsewhere.

  • Canva is less exposed immediately because much of its work involves physical posters, pictures and other outputs decoupled from software. Gamma is closer to the firing line because Claude already makes slides; the application can also import from Figma and export to Canva.

10. Design is collapsing into the same workspace as product and engineering

  • Jason contrasted the new workflow with his Adobe experience: public-facing design required a ticket, followed 60-90 days later by a PDF that product and engineering spent further months implementing. AI first let product teams prototype in Replit and Lovable, then commit directly alongside engineers.

  • Claude Design’s export into Claude Code closes the loop. Jason still wants humans for an exceptional homepage or splash screen, but teams shipping daily will use the integrated result now and improve it later: “No one’s saying it’s pixel perfect. Don’t have time.”

  • Rory compared the attack to Google Workspace gnawing at Microsoft Office. Google could spend $1 billion “yanking Microsoft’s chain” without needing to win; 15 years later Office remained roughly a $40 billion business, illustrating outcomes ranging from minor low-end user loss to 25% revenue impairment.

  • Bundling makes this instance more credible than forgotten GPT stores or plugins. Design is the front door to software, coding is AI’s enterprise motherlode, and a unified design-product-engineering tool has a “better together” advantage worth sustaining near Claude Code’s core.

11. Mega-growth funds monetize access even as consensus erodes alpha

  • Accel’s $4 billion leaders fund and Sequoia’s $7 billion growth fund confirm where institutional capital wants exposure. Rory agreed directionally: companies stay private longer and outcomes are larger, so capitalism is raising vehicles capable of writing the required checks.

  • His caveat was cyclical. Growth produces excess returns when valuations are depressed, as in 2022, or when investors possess an insight the market lacks, such as recognizing ChatGPT’s importance early; plentiful capital and universal belief in AI weaken both sources of advantage.

  • Jason argued the rounds can physically absorb the funds. At $30 billion or $122 billion, relationship-rich firms can secure allocations even as the fourth name in a release; repeated access to Dario or Sam turns brand and proximity into a scalable distribution advantage.

  • Harry called some Anthropic SPVs “the most egregious face ripping” he had seen, including 8% charged upfront. Sequoia’s opportunity is straightforward: LPs demand one scarce asset, the company values its name, and the fund can sit between them collecting economics.

12. Rippling proves high-growth SaaS survives, while APIs reshuffle fintech

  • Rippling reportedly crossed $1 billion ARR at 78% year-over-year growth after sitting below $500 million only 11 months earlier. Jason stressed that it is not merely growing fast but accelerating, calling Parker Conrad a candidate for CEO of the year.

  • Rory’s verdict on the category was blunt: “The whole SaaS is dead meme is bullshit.” Low-growth SaaS is unattractive; a billion-dollar company accelerating near 78% receives an excellent IPO because investors object to 10% growth and capped markets, not to subscription software itself.

  • Payroll is unusually durable because errors trigger immediate employee complaints and statutory or even criminal consequences. Customers do not want non-deterministic processes; they want a competent vendor accepting responsibility, making the core proposition largely orthogonal to AI.

  • Agents can still reorder adjacent winners. Jason’s planned AI VP of Finance will choose collections and card providers by API, not dashboard; after six years on Brex, SaaStr could switch to Ramp in one week if it wins the bake-off. Rory’s distinction: the stack remains, but API-forward vendors take share.

13. Salesforce’s headless move is a bid to govern the enterprise agent layer

  • Rory explained headless Salesforce as separating its two assets: the human UI and the database-workflow system holding leads, customers and pipeline. If agents replace seats, Salesforce can surrender the interface while retaining the back end those agents require.

  • Jason argued Salesforce has effectively been headless since launching its enterprise API in 2006. SaaStr’s agents already find it the best API in their stack and are using it to move Marketo data into Salesforce within days, despite the broader product rollout arriving at traditional B2B speed.

  • The larger bet is “agent fabric,” not simple orchestration: a trusted layer that knows every operation performed by hundreds of agents and sub-agents in real time, supplies context and guardrails, and handles governance, auditability and security before autonomous systems approve orders or alter records.

  • Rory’s analogy came from early e-commerce analytics. Retail executives once lost the ability to “walk the floor,” creating NetGenesis and Omniture; executives now need visibility into far more powerful, decision-making bots. Salesforce is well placed, but only if it can execute before its installed-base advantage expires.

14. AI M&A is trapped between unaffordable startups and instant obsolescence

  • Elad Gil’s advice was to discuss exits annually and, where possible, sell AI companies for an easy $1 billion. Jason agreed with the discipline but questioned the buyers: a public-company CEO told him startups with $5 million revenue routinely demand $1 billion after their latest rounds.

  • HubSpot, valued around $12 billion in the conversation, can risk roughly $50 million but cannot repeatedly spend $1 billion. Rory noted that among approximately 250 AEO/GEO competitors, an incumbent should still find an affordable, capable target somewhere below the first five.

  • Price is only half the problem. Jason argued HubSpot may have bought a competitive AEO product for $30 million, only to find it stale within months; an acquisition that once allowed a year-long rebuild now risks becoming irrelevant before integration finishes.

  • Rory sharpened the organizational diagnosis: the challenge is preserving urgency and speed after acquisition. With AI-native products potentially becoming obsolete within weeks, buying talent or code without retaining its operating tempo may destroy the reason for the deal.

15. Cerebras has earned a credible second attempt at the public market

  • Harry cited $510 million of 2025 revenue, 76% growth and $219 million in 2024, while noting that earlier G42 concentration concerns had been resolved. Rory corrected the profitability narrative: an accounting liability reversal created noise, but operations still lose money.

  • The venture achievement is a decade-long semiconductor build. Founded in 2016, Cerebras created a “big-ass wafer-scale chip” optimized for fast, low-latency AI inference in a market with few buyers and hyperscalers increasingly designing their own silicon.

  • Middle Eastern deployments and a cloud inference service proved the chip in use; one Rory-backed company found it genuinely fast. More recently, Cerebras signed deals with AWS and OpenAI; the OpenAI deal was cited at $20 billion of commitments, though Rory cautioned, “Who the hell knows what that means.”

  • Groq’s $20 billion sale creates one benchmark, but Cerebras is a high-beta public call option. With Nvidia around $5 trillion in the discussion, Rory observed that 1% equals $50 billion; when risk appetite is high, the only other standalone exposure can price far above fundamentals.

16. Selling versus staying independent is only obvious after momentum turns

  • Jason contrasted Groq’s $20 billion exit with Cerebras enduring a public-company roller coaster for another decade. Harry preferred the SalesLoft path — sell to private equity for $2 billion and “peace out” — because founders also optimize for time, risk and emotional cost.

  • Rory’s answer resisted a universal rule. Google was glad it did not sell early, Jensen is glad Nvidia stayed independent, and founders enjoying the work may rationally continue; “when it works, you’re glad you didn’t sell, and when it doesn’t work, you wish you had.”

  • Jason returned to Gil’s governance test: once a year, boards should ask honestly whether they are Nvidia or the company whose momentum has already broken. The danger is that every CEO imagines being Jensen while employee RSUs deteriorate and talent leaves for the rising platform.

17. Jensen Huang’s China argument exposed incompatible starting assumptions

  • Rory found Huang strongest on the operating facts: Nvidia sells chips to customers issuing purchase orders and manages a grounded relationship with TSMC. On Anthropic, Huang acknowledged he wished Nvidia had invested but said the company lacked capacity then for venture checks on today’s scale.

  • The China exchange with Dwarkesh failed because the parties never shared two priors: whether China is a normal commercial competitor or a strategic enemy, and whether frontier models are ordinary software or as dangerous as uranium. “Neither of the nouns in the sentence” was defined.

  • Huang was also unmistakably talking his book. Rory put roughly 30% of Nvidia’s market and $40 billion of revenue in China; after lobbying the president for permission to sell there, no question from a 25-year-old podcaster would make him abandon that position.

  • Dwarkesh still earned credit for pressing a $5 trillion-company CEO with 30 years of domain knowledge. Rory did not think Huang proved the argument, but he recognized a semiconductor executive who could “hold his corner” through an unusually difficult interview.

18. AI’s Bay Area concentration creates a narrow European counter-position

  • Jason cited Elon Musk’s reposted claim that 91% of AI unicorns are now in the Bay Area. Harry accepted that most leading minds want Silicon Valley and that power is recentralizing there, while pointing to DeepMind and ElevenLabs as proof that exceptional European companies remain possible.

  • His investment trade-off is supply. Silicon Valley combines 91% of the companies with 91% of the capital and ferocious competition; as a top-three European brand, Harry prefers a market with perhaps one-tenth the competitive intensity over fighting Benchmark, Founders Fund and Andreessen for every allocation.

  • Rory translated the 91% into an equilibrium: “Bay Area wins,” but congestion, recruiting cost and scarce talent leave some margin for Europe. The recent advantage was tribal knowledge circulating through San Francisco hacker houses after 2022; as that knowledge disperses, location should become less determinative.

Jason Lemkin

I'm not sure they're buying them for $60 billion.

Harry Stebbings

This is 20VC with me, Harry Stebbings. It's my favorite show of the week. Rory O'Driscoll and Jason Lemkin analyzing the biggest news in tech. Now, this week on the agenda, my God, breaking news, Cursor acquired by xAI or SpaceX, otherwise known, for $60 billion with a $10 billion break clause. Tim Cook announces he's stepping down from Apple after an epic run. Anthropic turns down $800 billion funding offers and crosses the trillion dollar mark on secondary markets. And then Anthropic launches Claude Design. As if it wasn't eating everyone else's lunch, now it's going after Figma, Adobe, Canva. This is an epic one. As always, we did not fall short of banging news to cover. But before we dive into the show today, let me tell you about Omni. It's an AI analytics platform, and it solves a problem every scaling company hits. Your team needs insights, not just data lookups, the stuff that really matters, and it's critical to get it right, like CAC payback periods and net dollar retention. For AI agents to act on your company data, they need your business context, your definitions, your logic, your permissions, and that's what Omni's governed context graph provides. Your data team defines it once, then anyone, your ops lead, your CFO, your PM, can ask a question in English and get an answer in seconds. Perplexity, Mercury, and DBT run on Omni. And 20VC listeners get a free three-week trial. Three-week, very specific. Not a month, but three weeks. Go to omni.co/20vc. That's omni.co/20vc. After Omni helps you find the right customers, Checkout helps you close them. Over the past 15 years, Guillaume Pozas has led checkout.com through what he calls the velocity years, a period of hypergrowth with relentless product building. The lesson? High growth is a gift, but it demands ruthless focus. As his mother put it, "Play the game you're good at." For checkout.com, that game is digital payments, obsessing over data, chasing basis points, and compounding learnings over time. And that discipline is paying off. 2025, checkout.com processed over 300 billion in total volume, up 64% year over year, and returned to full-year EBITDA profitability. They now support over 1,000 enterprise merchants globally, including 63 that process more than a billion annually with brands like eBay, Vinted, Amex, ASOS, and Temu. Guillaume's message though, it's pretty clear. They've earned the right to win anywhere. Now, they're investing in innovation across marketplaces, issuing financial experiences, and agentic commerce. If you want payments built for what's next, talk to the team at checkout.com. That's checkout.com. While Checkout powers the moment money changes hands, Invisible powers the people behind the work. Why don't we hear more real AI success stories from big companies? The models are insanely good, but implementation's the problem. It's really, really hard. There's data all over the place. There's legacy tech and manual workarounds. It's a Ferrari engine in a shopping cart. Meet Invisible. Invisible trains 80% of the top models and then adapts them to the messy reality of your business. Take the Charlotte Hornets NBA team. Invisible took years of game tape and analog scouting notes to go from uncertainty to a draft pick and Summer League championship win in weeks, not seasons. Get the data in order first, and suddenly AI can do almost anything for you in the enterprise. If you want AI that hits the P&L, go to invisibletech.ai/20vc.

Rory O'Driscoll

You have now arrived at your destination.

Jason Lemkin

I'm not sure they're buying them for $60 billion.

Rory O'Driscoll

If your stock is valued at 100 times revenues, you can buy things that are trading at 10 or 15 times revenue all fucking day long.

Jason Lemkin

I think there'll be a $100 billion deal in the next 12 months.

Rory O'Driscoll

I think this will stand as the high-water mark of private M&A for a decade.

Harry Stebbings

Boys, we recorded yesterday, and last night some very big news happened. I messaged this morning saying I think we should actually do a little segment this morning and discuss this because it is so notable. Last night, it was announced that Cursor was—Rory, you're going to correct me whatever I say, so I'm just going to do it—

Rory O'Driscoll

Yeah.

Harry Stebbings

—and you can correct me. Cursor is being acquired by xAI—SpaceX, in other words—for $60 billion, with a $10 billion break clause by the end of the year. Rory, please correct me on—

Rory O'Driscoll

It's too early, Eric. We shouldn't overthink it. Basically, they have an option to acquire Cursor, and if they don't, they pay $10 billion for the work, but it amounts to the same thing.

1. Cursor Gets A Sixty Billion Option

Jason Lemkin

Listen, it's very interesting because, first of all, this is an epic deal, right? If it happens—

Rory O'Driscoll

Yep.

Jason Lemkin

—it's $60 billion in about 3 years from seed funding. We thought Wiz was big, guys. When we started the show, we were talking about—

Rory O'Driscoll

Agreed.

Jason Lemkin

—Windsurf. That was $2.5 billion. Now it's $60 billion. These are numbers that we have not seen in 3 years from founding.

Now, having said that, we haven't read whatever the deal is, right? I'm not sure they're buying them for $60 billion. There's a lot of stuff going on between these companies. Senior engineers have already quit and gone to xAI. It appears that Cursor may be, at least notionally, the largest customer for Groq for coding.

Their market share has fallen, and they've already committed to using a lot of the 1 million GPUs that SpaceX—whatever we're calling it—has in Colossus. So there's a lot going on here, and what this call option is and how much it's tied to Cursor having enough of a state-of-the-art model that they're worth buying—because an IDE isn't enough—there's a lot up in the air—

Rory O'Driscoll

Yeah.

Jason Lemkin

—and it's probably why the deal is structured like this, beyond the fact that SpaceX is going public.

Rory O'Driscoll

Agreed.

Jason Lemkin

Which adds to the complexity. There are a lot of moving pieces here, so I wouldn't be surprised—

Rory O'Driscoll

Yeah.

Jason Lemkin

—if it doesn't close, because I think there is a big milestone that needs to be hit. Cursor needs to make progress, and maybe I'm wrong.

Rory O'Driscoll

Wow, such a lot to unpack, but I want to actually cover all this. First of all, you're right: this is the biggest private acquisition ever in venture. It used to be Wiz at $32 billion. Before that, it used to be WhatsApp at $16 billion—an individual private check before going public. Now it's a $60 billion outcome in 2 years.

You're right: if this deal closes in 3 months, this is the biggest privately held venture acquisition ever, period, full stop, end of story. It's an amazing result. So you're right, Jason. Before you get into the noise, that's the venture takeaway.

Jason Lemkin

Jesus.

Rory O'Driscoll

And wow. We'll come back to who's making money—

Jason Lemkin

And it didn't even take 20 years, like most of our funds take. This will be 3.5 years.

Rory O'Driscoll

No, absolutely. The big picture here, guys, actually makes sense. I was thinking last night of doing a little cute tweet before I got your note, Harry. It's like you've got these guys called Cursor walking down the street saying, "Hey, good news: we've got an exploding business, a couple of billion dollars in ARR. But bad news: we have shitty gross margins because we need our own model and we need compute." And Elon goes, "Hmm, hold that thought."

This is a guy walking around with a whole bunch of compute, a reasonably good model, and literally no revenue. It's like a marriage made in heaven. Because, look, let's get real here: it's one of those deals that makes sense for both sides, which is why I think it'll close. That's different from saying it makes full financial sense. I'll come back to that.

If you think about it, from SpaceX's perspective, all the narrative has been, "SpaceX is amazing, Starlink is amazing, WTF is with this xAI? Why did they stick it in Twitter and xAI? I know Elon loves AI, but it makes my head hurt." The truth is, they've spent $20 billion, plus or minus, on the most amazing data center—the Colossus data center—with literally hundreds of thousands of GPUs, but they're not doing a very good job of selling them, so they have relatively little business. All other things being equal, that's a bad place to be.

Now, capacity is scarce, so I'm sure they could have found white-label work selling to one of the hyperscalers. But instead, they found this company that's in the motherlode—the mother of all AI markets, which is coding—that has customers and revenue, but has shitty gross margins because it doesn't yet have a full standalone model and massively needs compute.

So they put this on top of their thing, and the vertically integrated thing just looks a lot more attractive. You literally have a company that's making $3 billion in revenue and spending $3 billion on gross margin, and you have another company that's burning $18 billion. You put them together and the $3 billion gets canceled, and at least they have some revenue.

It's still not enough to cover the xAI nut unless they can grow this thing, but at least it looks like a business now. I think what's really clever from SpaceX's perspective is that they couldn't do it before the IPO, because you can't close that deal. It delays everything. It's actually a very clever structure.

So they're basically going to be able to say to the public, "You love Starlink and you love our space business, and you didn't like this other thing, but we've got a plan to fix it. Once we close at $2 trillion, we'll exercise our option, we'll buy this thing, and suddenly we'll be a $4 billion- or $5 billion-revenue AI frontier lab. My Claude Code equivalent is called Cursor, and I actually have the rest of the stack. There, I told you I'd make a good business out of this."

Harry Stebbings

Which is an important note just for people to understand.

This is not a transaction closing today. This is in 6 months, post going public.

Jason Lemkin

Yeah.

Jason Lemkin

And it's a very strategic way for them to do it because they can't do it now—

Yeah.

Jason Lemkin

—just before going public.

Rory O'Driscoll

And that's why I think of it as an option. You're right, Jason, there's been some commentary that said maybe this option is in part Elon doing a try-before-you-buy. Let's see if the model works on our stuff and vice versa, and if it doesn't, we'll just pay you $10 billion and walk away.

I don't know. I'm not going to guess, and there's probably some of that. But fundamentally, it's costing you nothing today because you're not writing the check today. What Elon is basically saying is, "If my IPO happens, and it probably should, and I want to buy this thing, I will because I can. And if it doesn't, I'm promising $10 million that I don't plan to have to give unless, God forbid, the IPO doesn't happen."

So he kind of mentally says, "I haven't fixed this story today, but I have a plan to fix it. You can see the plan to fix it. So stop bugging me in the roadshow about xAI. Let's talk about rockets in space."

I think—listen, Elon walked away from the Twitter acquisition, so he'll walk away from a bad deal, don't get me wrong, if things change. But you don't promote it the way SpaceX did all over X—

Jason Lemkin

Yeah.

Rory O'Driscoll

—if you're not basically saying this deal's going to close, right? This isn't a change. Otherwise, if it's just vendor financing with an option to buy, you bury it.

Jason Lemkin

Yeah.

Rory O'Driscoll

You bury it—

Jason Lemkin

And—

Rory O'Driscoll

—because it's round-trip revenue—

Jason Lemkin

Bury it.

Rory O'Driscoll

—so someone will use—someone will use our Colossus for model training, right? You'd hide it if it was—

Jason Lemkin

I agree.

Rory O'Driscoll

—just a vendor deal, right?

Jason Lemkin

Agreed. Even though I don't think the whole business effort here makes sense, right—trying to enter this market, trying to be the third or fourth foundation-model player as xAI/SpaceX—I don't think it made any sense.

But once you've decided to do that, this deal makes sense at the margin, because you've incurred all the negative parts of being a subscale hyperscaler with $20 billion in burn, and now at least you have a business. So I'm with you, Jason. I think they're pushing this because they're like, "Oh, this makes sense, and actually solves a problem for us."

It obviously solves a problem for Cursor, makes the P&L of their business look very different, and therefore it's a win for both sides and it closes.

Harry Stebbings

If I push you, who is getting a better deal? They're expected to finish the year at $6 billion, so this is only 10x end-of-year revenue.

Rory O'Driscoll

Seriously? We just talked on this pod before about Figma at $20 billion and about Groq to NVIDIA—you just said Groq to NVIDIA for $20 billion—

Jason Lemkin

$20 billion.

Rory O'Driscoll

—and getting to be—

Jason Lemkin

That's it.

Rory O'Driscoll

—not having to be a public-company CEO is the—

Jason Lemkin

Again—

Rory O'Driscoll

—second-best gift to Salesloft selling for $2 billion. This is a gift. They don't have to do—

Jason Lemkin

Mm.

Rory O'Driscoll

—the hard work. They've only been doing this for 3 years.

Jason Lemkin

Totally.

Rory O'Driscoll

They've only—

Jason Lemkin

I—

Rory O'Driscoll

—been doing this for 3 years. I was watching—

Jason Lemkin

I—

Rory O'Driscoll

—some images of—what's Michael's last name, the CEO?

Jason Lemkin

Tru.

Rory O'Driscoll

Okay. I was watching something on Twitter with him and Garry Tan a year ago, okay? Garry Tan's got a big job, right? Garry Tan looks like he's aged 8 years since the video. Michael looks the same.

This is the time to sell, because you don't—listen, this is maybe a little meta, but what I've learned as a founder is you don't get—it doesn't hit you until around year 4 or 5. This is just the way humans are, okay? And he's young, but look at founders you've invested in or met with.

Around year 4 or 5, the weight's there, okay? You can see it in the bags under their eyes, whether they're 21 or 61, and you've got to double down.

Jason Lemkin

No.

Rory O'Driscoll

At 4 or 5 years old, you've got to sit around and say, "Do I want to sign up for another tour of duty?" These guys don't even have to do a second—

Jason Lemkin

Yeah.

Rory O'Driscoll

—tour of duty. This is a pretty good deal, Harry. If you're 10 or 15 years in, you know what to do. 3 years in, you go, you've got to go home.

Jason Lemkin

There's no—guys—

Rory O'Driscoll

You've got to go home after 3 years for $60 billion. 3 years.

Jason Lemkin

Can I—

Rory O'Driscoll

3 years after $60 billion.

Jason Lemkin

Can I help you on this? You asked which of the 2 is getting a great deal, and the answer is they're both getting a great deal.

Harry Stebbings

You've got to give 1 name.

Jason Lemkin

No, I'll tell you who's losing. There are 2 reasons they're getting a great deal. Everyone's getting a great deal.

The first is it actually industrially makes sense. The 2 things go together well, right? So there's some business logic. Whether or not the SpaceX-Elon team can manage these kinds of researchers over time, that's TBD, given how the X story has unfolded so far. But it makes industrial sense for these 2 companies to come together.

Why are they both getting a great deal? You're missing the point, because there's a third player in the game here that you're not mentioning: the future SpaceX public shareholders. They're valuing a $20 million-revenue company at $2 trillion, which is 100 times revenues.

If your stock is valued at 100 times revenues, you can buy things that are trading at 10 or 15 times revenue all fucking day long. As long as SpaceX is worth $2 trillion, if Elon wants to scratch an itch to clean up a subsidiary of SpaceX that isn't quite working out, and he can chuck $6 billion on the table, for context here, it's 3% of SpaceX's alleged market cap in return for 15% to maybe 20% of their total revenues.

Rory O'Driscoll

It's not even material.

Jason Lemkin

Yeah. When you have a high-priced stock, you can buy any pretty things that you want. And what happened—

Rory O'Driscoll

And you should.

Jason Lemkin

And you should.

Rory O'Driscoll

You should—

Jason Lemkin

Well, yeah—

Rory O'Driscoll

—because it may not last.

Jason Lemkin

You might.

Rory O'Driscoll

Right?

Jason Lemkin

Yeah, exactly.

Rory O'Driscoll

This is the opposite. When we did the other version of this pod, we were talking about how so many public software companies, if you're now trading at $10 billion, you can't even spend $1 billion to buy a couple of kids—

Jason Lemkin

Yeah.

Jason Lemkin

As long as SpaceX is worth that kind of multiple, it's revenue-accretive, story-accretive. And another reason it makes sense to Cursor is the number of people who can write a $60 billion check for a roughly break-even gross-margin business can be counted on the fingers of 1 hand, removing at least 1 or 2 of them, right?

Rory O'Driscoll

Yeah.

Jason Lemkin

Of those acquirers, most of them, in most situations, would be unable to do so because of the DOJ. This is the only game in town.

Rory O'Driscoll

It helps to be founder-led, too, to write a check like this.

Jason Lemkin

Yeah. Oh, totally. Yes. This is a—

Rory O'Driscoll

If you're going to Google, again, you're like, "I know we did Wiz at $32 billion and it's crushing it, but we want to buy Cursor for twice that?"

Jason Lemkin

Yeah, yeah.

Rory O'Driscoll

I mean, it's a tense board discussion, right?

Harry Stebbings

If I'm an investor in the business, what does it mean for me? When do I get my cash? What are my LPs getting back?

Jason Lemkin

You're getting a lot of money. I mean, look, you're getting your cash if this deal closes, I think, in 6 months, which is the back end of this year, assuming the IPO happens in June. You're ecstatic. It means—

Harry Stebbings

And you're getting cash, or you're getting SpaceX stock?

Jason Lemkin

No, well, you're almost certainly getting stock. The IPO is $75 billion. So then the second-order questions are: Is it registered? If you're getting that stock in the restricted period, is that going to be restricted too?

Because the whole dynamics of the SpaceX lockup, given the size of the round, are going to be 1 of the most interesting parts of the whole transaction. There's talk of, like you say, employees getting out early. There's also, to be fair, talk about some shareholders being held in for much longer, because the combination of a $2 trillion—I mean, it's going to make a difficult problem around float management slightly more complex, because a $75 billion raise on a $2 trillion valuation is a teeny, tiny float.

And going from there to the other 97% of the company being freely tradable in 6 months just doesn't work. So I'm sure there's going to be a convoluted float-management thing, and then this stock will probably go into that.

So I wouldn't be surprised to discover that, as investors in Cursor, congratulations, you are now investors in SpaceX. You've got that fluctuation risk for 6 months, plus or minus, but on the other hand, you are getting $60 billion.

Harry Stebbings

Shut up.

Rory O'Driscoll

Do we know if the $2 billion at $50 billion happened—that round happened last month?

Jason Lemkin

I don't.

Rory O'Driscoll

We don't.

Jason Lemkin

I don't know.

Rory O'Driscoll

And then do we think it would be—

Harry Stebbings

Well, they're tied together. I don't know all the machinations, but—

Jason Lemkin

I think I saw one—

Harry Stebbings

They're tied in some fashion.

Jason Lemkin

I saw one that said—Can I stop you there?

Harry Stebbings

Yeah.

Jason Lemkin

I think I saw one that said it's not happening because this obviates the need for it. If you think about it, they probably had reasonable—I’m just winging it here—but they probably had pretty good cash on the balance sheet, but they knew they had to spend a lot on compute.

And now one of two things is going to happen in September or October—six months. A, you're going to get $60 billion, and you have all the compute you'll ever need from Elon. And B, you're going to get $10 billion in cash, in which case you can buy some compute then.

So they can probably do some forward contracts. If the round hasn't closed, it probably won't close now, because there's no point to it and no one would want that stock. That's basically like buying SpaceX, and you'd prefer to just wait for the IPO.

Rory O'Driscoll

If we get stock back and it's locked up for 12 to 18 months, or whatever that is, how am I feeling as an investor? Am I still ecstatic? Again, if you're in the Figma IPO and you're locked up for 12 to 18 months, that's been brutal.

Jason Lemkin

Well—

Harry Stebbings

Mark wrote a blank check to Elon for the take-private of Twitter. Of course you're going to go long on Elon now, right?

Jason Lemkin

I mean—

Harry Stebbings

You're doubling down—

Jason Lemkin

Yeah.

Harry Stebbings

—on the same founder. It's not even—I don't know the cap tables, but for Andreessen, it's almost like merging 2 portfolio companies at a meta level. Maybe not actually—that's maybe not going to be accurate, but intellectually, it's like that.

Jason Lemkin

No.

Harry Stebbings

Right?

Jason Lemkin

But, Harry, you are right. This is where the beauty of being in early at a low base is. I looked, and if you're in at the B, which is where—the A was Andreessen and Thrive; the B was Thrive-led. Basically, the C is roughly a 5X, the B is a 20X. I can't even calculate the A, but it's huge. And then the most recent round is a 2X.

If you're into SpaceX at $2 trillion, and that's only a 2X, you're scared, because there's a lot of volatility in that stock. That's the thing. You could end up breaking even—or go down. IPOs do go down, apparently, at times.

Facebook went down, plus or minus, I think, 40% from its IPO price in the first 6 months before 10Xing subsequently, just as a reminder. So you have risk in that deal. On the other hand, if you're sitting there at 20 times, in the case of the B, or 50 or 60 times in the case of the A, you're like, “Oh, well, if my 80X becomes a 40X, it's still a 40X. I'm fine.” So you can roll that dice.

Harry Stebbings

It's almost a defensive move if you're Andreessen. You've added value to your net asset value by combining these on paper, right? You're shareholders in both. You have an IPO; it has risk. You threw this crazy asset in there that's very immature, where the debt is exploding, right? Your balance sheet has been blown out.

Jason Lemkin

Yeah.

Harry Stebbings

How can I de-risk my IPO for a couple percent dilution? Done.

Jason Lemkin

I think you're correct in your analysis, though my comment is that the entire sentence implies that Elon consulted with anyone. My sense is Elon solved Elon's problem, and everyone else is along for the ride. They'll get the shareholder consent and the DocuSign, and they'll say, “Thank you very much,” and sign.

But you're right. If I owned a lot of SpaceX, oddly enough, to say it succinctly, I'd have been pissed at the xAI merger, because I didn't need that dilution. But having done the xAI merger, I'd be like, “Clever deal with Cursor, dude.” You had this negative thing; now at least you've got a full-on, full-stack story.

Is it the best story ever? No. On a standalone foundation-model-lab bet, would you prefer to own Anthropic rather than xAI plus Cursor plus the Frankenstein of Twitter? Yeah, you would. But at least it's a full-stack story now.

This is a clever deal, again enabled by the 100X valuation you have—or at least allegedly have, but at least 50X. Even at the low end, if you play it out here, even if SpaceX trades, shock horror, at $1 trillion, which is a 50X revenue multiple, and even if Cursor gets to $6 billion by year-end, so it's a 10X revenue multiple, SpaceX is still buying stuff at 10X with stock that's trading at 50X. It's a great currency.

Harry Stebbings

Those arbitrages are real. You really should—

Rory O'Driscoll

Mm.

Harry Stebbings

—do a deal or two in those—

Rory O'Driscoll

No, no.

Harry Stebbings

—moments in time—

Rory O'Driscoll

No.

Harry Stebbings

—because they don't last. One way or the other, they don't last.

Rory O'Driscoll

No.

Harry Stebbings

Right?

Rory O'Driscoll

At the risk of stating the obvious, there is a lot more fucking differentiation in putting rockets in the sky than there is in forking an open-source IDE and running it on top of Claude Code.

I think the folks at SpaceX have done a much harder task, so they deserve a higher revenue multiple. Nonetheless, it's why you can write the check with such gay abandon.

Harry Stebbings

And look, if the next deal's at $120 billion, all is good.

Rory O'Driscoll

Going to your point earlier, though, Rory, of comparing it to Wiz, the other thing that's just amazing is the scale. It's not like a little bit more. It's double Wiz, give or take. I mean, literally, $32 billion versus $60 billion.

I was thinking about this this morning. For arguably a much less defensible business, I think the truth is this: It's why venture capital is amazing. It's the aha—don't play small ball.

If you play in the big markets, if you lean into the risk in the part of the cycle where the big guys want to win and want to catch up and want to be relevant in the space, then this kind of exit can happen, even if the fundamentals aren't great.

And on gross margin, Sequoia, who you might think are the doyens of the whole industry, always say, “We hate negative gross-margin businesses,” right? This is a break-even gross-margin business. But it worked, because the overarching comment is that AI is the biggest story of the last 5 years, and big tech wants to be a player. Cursor's a player in the biggest markets. Don't overthink it.

It's like the SaaS motto: “Who dares wins.” Those guys dared and they won. Now, the minute the tide goes out, we saw this in '99 too, once the tide goes out and people are scared of dot-com or scared of AI, all these high-burn, high-growth companies can get a real hit.

But what you're selling now, what they were selling, is a ticket to matter. And xAI won the ticket. So, to your comment on all the other guys, the Factories of this world, there are 10 or 20 other companies that are staring at irrelevance over the next 3 to 5 years, because they're all part of the old software development lifecycle.

Maybe they can't pay $60 billion, but they're all going to have to buy themselves or build themselves into relevance. So I think a lot of these can still have decent exits. The lad's right; you should be thinking about this. But I'd prefer to be in something like Factory, leaning in, than in some old-school code development tool that doesn't matter anymore.

2. The Next Hundred Billion Deal

Harry Stebbings

Another way to flip it around: on the one hand, sure, you should sell. On the other hand, there are 6 leaders with market caps above $2 trillion, right?

Rory O'Driscoll

Yeah.

Harry Stebbings

And we're adding SpaceX, so there'll be 7, maybe, above $2 trillion. They can all afford to pay $100 billion—5% of their market cap—or split it with some cash to not fall behind.

But Jensen's stressed about Tanium.

Rory O'Driscoll

Yeah.

Harry Stebbings

He can spend $100 billion, right? He already did a small deal, a $20 billion tuck-in. Nvidia, Apple, Meta, Amazon—

Rory O'Driscoll

It'll change, but—

Harry Stebbings

—Alphabet, Microsoft—all can buy a startup for $100 billion.

I remember back in the day at SaaStr Annual, Ryan Smith came and Qualtrics had just been acquired for $7 billion, and our jaws dropped back then. That was a lot for a software company. I asked Ryan, “Will this ever happen again?”

He's like, “I don't see why there won't be one for $14 billion next month. It's just getting good.” And he was right. The party kept going, to Rory's point, until it ended.

I think there'll be a $100 billion deal in the next 12 months. Which one it is, I don't know, but there'll be a $100 billion deal. This is going to be one of 3. That's my prediction. And I may be wrong, but there are 7 folks that can do it—do $100 billion.

Rory O'Driscoll

I'll take the under.

I think this will stand as the high-water mark of private M&A for a decade.

Jason Lemkin

Well, this clip is going to embarrass you, Rory.

Rory O'Driscoll

Because I just think it's anomalous in the sense that it's a combination of the absolute amount and the revenue multiple. The number of people who can write a check for less than 10% of their market cap where that's greater than $60 billion is, by definition, people above $600 billion. There are only 8 or 10 companies above $1 trillion, so it's a very finite group of people who can do those kinds of deals.

Everyone but Elon trades at under 10X revenues, plus or minus—yeah, it's pretty much minus. So they're not going to do a 60X. With the exception of Elon, because he can overpay because he's trading at 100X, or at least—

Jason Lemkin

Yeah.

Rory O'Driscoll

—plans to be trading at 100X. I wanted to spend a little time on that in a second. Everyone else trades sub-10, so you're not going to buy anything greater than 10X, which means to pay $100 billion, you have to buy something worth $10 billion, right?

As I did that, I realized, “Hmm.” At that point, a privately held company has to be doing, let's round down again, $5 billion to $10 billion a year to be worth $50 billion to $100 billion, to meet the Jason criteria of being bigger than this. If JPMorgan wanted to buy Stripe, you'd actually be right.

Jason Lemkin

Yep.

Rory O'Driscoll

So I can imagine a few circumstances where that could happen. And then you also have the DOJ stuff, which is obviously suspended for a while, but there you go. So you're not impossibly wrong, Jason, but I just think this is such a special set of circumstances.

It's a combination of a couple of things. One is it's a company trading at an amazing price, feeling the imperative to execute a transaction in an amazing space where there's a company willing to transact like that. The other thing I was thinking about is one of the bigger risks in this is that Elon is so risk-on. I would use the expression in poker when you're on tilt, when maybe it's not going well and your response is to double down and double down, right?

If you look at the whole Twitter/X transaction, unlike SpaceX, which has been a work of genius from day 1, it's been: “I bought Twitter. I didn't want to do it. Shit, now I have to do it. That's $44 billion. Well, if I'm going to do that, then I'm going to want to do the AI thing and spend $20 billion on CapEx there. Got no revenue. Let's put them together. Ooh, that's not working. Let's flip them into SpaceX for $250 billion. It feels notional, but whatever. That's not great, so let's buy Cursor for another $60 billion.”

He's basically going to keep doubling down until he wins in AI. It's astonishing; it's terrifying if you're a stakeholder, but that's the game he's playing and it's his company.

Jason Lemkin

Let me just tell you why you're wrong about the $100 billion deal, Rory.

Rory O'Driscoll

Okay.

Jason Lemkin

Just process-wise, okay? And I think your math suggests I might be wrong. I hadn't even thought about JPMorgan–Stripe. That literally could happen tomorrow. We could open up our X accounts. It would make some sense, right?

Rory O'Driscoll

Yeah. Wouldn't be crazy.

Jason Lemkin

Here's what happens when times are good but stressful, which is what's happening with all the leaders. Microsoft, Google.

Rory O'Driscoll

Hmm.

Jason Lemkin

The one thing I have been that I don't think you guys have been—I know Harry has been—is a senior VP on the other side at a big tech company during moments of change.

Rory O'Driscoll

Sure.

Jason Lemkin

And what we talked about all the time is, who the hell can we buy to get ahead? I will tell you, in all of these boardrooms, especially Zuck—you could already imagine in your head—they are saying, “I want to buy something that's going to actually effing move the needle for us. We'll do some tuck-ins. You guys can all have a chip. You can go buy a little thing for $50 million, $500 million, maybe even $1 billion. But I want something that will move the needle in AI now.”

Rory O'Driscoll

Yeah.

Jason Lemkin

“Bring me your candidates,” and they'll all be debated, but if it is above the line, those deals will get done for 5% or less of your market cap. I guarantee you, the discussions—

Rory O'Driscoll

Jason—

Jason Lemkin

Because I was there. It happens every week.

Rory O'Driscoll

Jason, even though I disagree on the number, I 100% agree on the description. That is exactly what happens. You wake up as the leader of this big company. You've got your $2 trillion market cap to defend. There's a technical thing that could make you obsolete. If you can spend 5% and deal with that, you do it. So I totally agree.

A large part of why venture makes money is that every once in a while you find yourself with companies that are kind of in the to-do list for corporate America, and they just write huge checks. So I do agree with you. I was pushing back on the $100 billion, but do I think it's kind of like the Overton window? A $10 billion or $20 billion acquisition from Zuck tomorrow morning—people wouldn't even blink. They'd be like, “Yeah, of course he's going to do that.”

So I do agree. I was just being, as you know me, annoyingly precise about $100 billion. But I agree with you. Every other CEO is looking at this going, “Hmm, what do we do now in coding?” I'm—

Jason Lemkin

And the other crazy thing that happens—I know it sounds crazy, but it is true—is on the other side, on the acquirer side, big company, right?

Rory O'Driscoll

Mm.

Jason Lemkin

Once certain numbers are breached, it doesn't mean you also—

Rory O'Driscoll

Yeah.

Jason Lemkin

—get $60 billion. It does not—

Rory O'Driscoll

Yeah. Overton window. Yeah.

Jason Lemkin

—like, you can't walk into a meeting and say, “I'm better than Cursor,” but it does psychologically change the breakpoint for what it—

Rory O'Driscoll

Agreed.

Jason Lemkin

—because when people go into deal mode in big companies—

Rory O'Driscoll

Totally. It happens.

Jason Lemkin

—as long as they can afford it, if they identified—

Rory O'Driscoll

Yeah.

Jason Lemkin

—the perfect target—like, remember when Benioff was willing to give his next firstborn to buy LinkedIn? He tried everything he could.

Rory O'Driscoll

Everything.

Jason Lemkin

Every share, every piece of debt. He couldn't get $30 billion to buy LinkedIn. That's a $100 billion deal today.

Rory O'Driscoll

You're exactly right.

Jason Lemkin

Benioff would—if Benioff could get $100 billion to buy something to change the face of Salesforce, the LinkedIn of AI, he would do it tonight, I guarantee you. Because he already tried to do it with LinkedIn. He would do $100 billion if he had it. All these guys have $100 billion.

So you go into a meeting, and I saw it at Adobe as Creative Cloud took off. People's views of price changed during my tenure as a VP. As soon as you see a Cursor deal, we don't give a shit. We're not the founders. It's not our money. It's not even our dilution. We don't care. What does it take to buy the next Cursor? $68 billion? Done. Boys on the corp dev team—

Rory O'Driscoll

And—

Jason Lemkin

—Stebbings and Lemkin on the corp dev team—

Rory O'Driscoll

Look.

Jason Lemkin

—close the damn deal.

Rory O'Driscoll

When you're fighting for relevance and defending an existing business, as you say, and even if I don't think they get to $100 billion, the fact that someone's done something at $60 billion means something at $10 billion or $15 billion will seem like, “Yeah, that seems like a good idea. Let's get it done.” It's expanded the Overton window of what's doable in M&A.

Last thing to say is this: great outcome for the VCs. Great outcome for—I mean, OpenAI. Just as a reminder to everyone, Sam Bankman-Fried did the seed. Poor Sam. The liquidator did his job and sold it for a 1X that could have been a couple hundred X.

OpenAI did the A, I think. Actually, OpenAI did another seed. Andreessen and Thrive did the A, then Thrive led the B and doubled down the whole way through. I don't know what the returns are on the A, but on the rest of it, I think they put in $800 million and got roughly a 4X.

Great outcome for everyone. It's a great venture story. All credit to the founders. This is a pretty good 3-year run. The 2 best deals that Thrive did amazingly are this one and OpenAI. The aggregate return on OpenAI, I think they're at a 4X, maybe 5.2X—I can't remember which. And on this, excluding the A, I saw the announcement, but it excludes the A, which is bullshit, because the A is going to be a 50X. On the rest of the money, it's a 4X.

What it says is there are small numbers of growth funds that can deploy masses of capital, hundreds of millions of dollars, and get a high-end venture return of a 4X, which is a good return. If you were doing an A or a B, that would be the high end of a good outcome, not yet great. You want a 10X-plus on your best stuff. But the fact that you can do it on $1 billion is what makes this special.

It's not that you're going to outperform some little seed like Neo. Again, it's the classic thing, Harry. It's actually why, even though you think it's simple to one-dimensionally measure venture returns, it's actually not. If you only have $1 to invest, and you can give it to Neo and get a 16X, or you can give it to Thrive Growth and get a 5X, you give it to Neo, right? From that perspective, you just rank on pure return. But if you're sitting there with $2 billion to invest, it doesn't matter what you do with Neo. And Thrive—

Jason Lemkin

Yeah, it doesn't matter.

Rory O'Driscoll

They have proven they can take $2 billion and turn it into $6 billion, and you're giving that product all day, every day. From a personal net worth perspective, if you run the math as being the lead GP on Thrive, it turns out to be remarkably good business.

We'll probably find that these products aren't even the same products anymore.

No.

Rory O'Driscoll

Okay?

Absolutely. No, for sure.

Because they were never the same product. But if I can put $5 million, $8 million, or $10 million into an emerging GP, and they don't expand into a multibillion-dollar fund—

Yeah, yeah.

—it is irrelevant to a large endowment, unless you've got a little chunk of guys who are just doing emerging managers, and they'll put in the work or they'll put in the time, or you're a—

Yeah.

Rory O'Driscoll

—smaller LP. And as the numbers get bigger, it just doesn't matter making 10X on your Neo if it's irrelevant to—

Yeah.

Rory O'Driscoll

—your portfolio, right? It's irrelevant.

Different—

It's irrelevant.

Jason, different products for different folks. But I just gotta call out the last comment I'll make on Thrive: if, quote, all they'd done was the late-stage stuff and gotten that 4 or 5X blended, that would be wildly impressive. But the fact that, on top of that, they were in the A alongside Andreessen is almost proof. The typical rap on only doing late is that if you're only doing late, you can't get the early round.

It's very impressive to have shown up for the A and then stuff the money in like foie gras from the B and beyond. That's really the best way to play that growth stage. If you have enough ground game to get in even under the tent for the A, and then you can stuff $1 billion in from there on in, that's the way you get the best of both worlds.

You've got some action on the table with a 40X or a 50X, is my guess, on $10 million. Yay, you. Maybe $20 million. Yay, you. And then you've got all the action in the world on the table of $800 million at an aggregate of a 4 or 5X, and that feels great. Mind you, you will be sweating that SpaceX lockup. You will know the day and the hour when it expires.

3. Tim Cook Exits Apple

Harry Stebbings

All right, boys. On to the next bit of news. Tim Cook announces he's stepping down from Apple after one of the greatest runs, from $350 billion to $4 trillion. John Ternus will take over as the new CEO. Twenty-four years at Apple for him. He's been in hardware before.

How do we read this news, boys? Is this what we expected in terms of replacement? How do we feel?

Rory O'Driscoll

On the one hand, he's 65, right? Obviously, the board has been talking about this for years. They found their successor inside. They didn't go through the chaos of Shantanu resigning from Adobe when they still don't have a CEO, right? They did it all the right way. They did it at the perfect age, optically.

But the big question, along with Reed Hastings and Shantanu and others, is: folks turn over every year, and I'm sure Rory's going to say that, but are they all leaving because they don't have an AI strategy? I don't think Netflix has one. I don't think Apple has one. I don't think Adobe has a great one, even though they have AI. I actually learned this week that AI in Illustrator is great. AI in Adobe Illustrator is a 10, but it hasn't changed the trajectory of the company, right?

So what would you do if you take an Apple from, whatever, 12X its market cap, successfully chaperoned the Jobs agenda to epic heights, and AI is changing the world? I mean, this show gets stale in less than a week. We've got to do four of these a week. I just don't know if this is normal retirement or, deep down, folks are like, “Jesus, I'm not up for this.”

Because I think most CEOs I work with are not up for it. Most pre-AI CEOs are not up for it. You guys know this too from your portfolio. Not all, but most are not actually up for the massive amount of—what's more than 996? 12-12-8? I don't know what it is—but most humans aren't up for it.

I think, as a general comment on AI terror, for lack of a better word, and the feeling that you need to be on top of it as a CEO, you're right. On the specifics of Cook, I think he's been operationally excellent the whole way through. He was operationally excellent taking over the job. He's run it excellently for 15 years, and I think he's operationally excellent on exit.

The proof of that is the stock barely budged. The stock is not always right. There are some fun examples of people thinking positive things in the short term and then being totally wrong. But the fact is that he resigns, not unexpectedly in the aggregate given his age of 65, but unexpectedly in terms of media timing, and the stock moves less than 0.5%.

That basically says organized, well-managed transition, and I think he was smart. I think you get out on a high. The recent quarter's great. As you say, 3 or 4X the revenue, 3 or 4X the revenue in 15 years, 3 or 4X the operating income. It's 12X the market cap, and just as a reminder, because they buy back shares, it's 20X the stock price.

As a holder of the stock throughout that entire period, God bless you. Just a great run. And handing over an internal successor, which, as you say, is proof that you had your shit together and you had a couple of options, and they went with this option. One of the other guys had pulled back recently.

I think getting out on a high is not to be underestimated at 65. I think it's a home run the whole way through. You just give credit where credit is due. It's one of the great runs.

Second comment: you're right, there is still some pending question: do you have an AI strategy? But honestly, Jason, I'm going to commit heresy now. If you look at the 3 companies you named—Netflix, Apple, and Adobe—I would argue that Adobe has to have an AI strategy. It's existential.

Apple is somewhere in the middle. Yes, it's pathetic that Siri is not good, but you can continue to build the hardware platform that everyone gets AI on, I think, for a long time to come. I don't think they're going to be replaced by the OpenAI mobile phone anytime soon.

And Netflix, I think it's the least AI-centric problem out there. They have interesting challenges. They didn't get that big acquisition done. They're a media company. They have media challenges, as Benedict Evans says so well. So it's in the middle in terms of AI terror.

Jason Lemkin

Well, look, for what it's worth, on Netflix, Blockbuster peaked in 2004 with 994 stores. Six years later, it went bankrupt. I think what happened with Netflix—and I'll tie this together to a macro theme that Harry brought up on Twitter and others—I think all that has to happen is for these leaders to be maimed or for there to be stealth churn.

I'll give you an example. I've paid for Netflix—I'm embarrassed to say how many years I've paid for Netflix—since it came on flash drives or CD-ROMs. I've been paying forever. Its price has gone up, but it's not worth my while to churn, right?

But I watch so much YouTube, and on YouTube it's becoming more and more AI-generated content. I've stealth-churned off Netflix. I'm still paying. I may pay for another year, but as more and more folks stealth-churn off Canva because they're using AI tools, as we stealth-churn, now that we've moved to Claude, Amelia hasn't used OpenAI in 4 months.

So we're paying for OpenAI because it's $100 a month, but she's stealth-churned. My caution—and I think this is why so many leaders are at risk—is that stealth churn is everywhere.

I used to think, as a B2B guy, that MAUs, WAUs, and DAUs were the dumbest metric. I'd get an investor update with, “Hooray, our WAUs are up,” and I'd be like, “I know Mamoon used to cheer that at Slack, but that's a sign of a struggling startup.” I wanted to see the revenue. Show me the money.

Now I want to see the MAUs and WAUs. I want to see your usage of Illustrator, Figma Make, Netflix, and Canva. I want to see those going up faster than revenue. This, I think, is the ultimate B2B test, or anything: are your MAUs, WAUs, and DAUs growing faster than revenue?

If so, things are probably working for you in the AI age. If they're declining, just like net new customer count, right? If it's declining, you're hiding. And I think Netflix just bought Ben Affleck's startup for like $300 million, right? Or something like that, Harry. They're well aware that folks are watching AI-generated Star Wars content on YouTube and not watching Netflix.

So, I don't know. I would step down, too. I just might.

4. Anthropic Reaches One Trillion

Harry Stebbings

Speaking of Amelia not having used OpenAI in 4 months—

Jason Lemkin

Yeah.

Harry Stebbings

—so deeply entrenched in her workflows with Claude, Anthropic turns down $800 billion funding offers. I just came back from an LP literally 10 minutes ago, and I said, “What's your biggest challenge?”

They manage the money for some of the largest families in Europe. And they said, “All of our families just want one thing: Anthropic. And the challenge we have is they don't want anything else, but they all want Anthropic, and that's the only thing they want.”

Every dollar wants a home in Anthropic right now, and it's caused secondary-market prices to surge to $1 trillion for Anthropic. The implication is that there is a clear market belief that they have won the enterprise race, and they've surpassed OpenAI now at $850 billion with their latest valuation.

How do we read that? How do we reflect on that?

Jason Lemkin

The fun thing in all of this—and I put “fun” in air quotes—is that it’s so fluid. Sam just said Codex usage is up 50% in 1 month. As we’re taping this, OpenAI is going to launch autonomous agents: OpenAI on steroids, 24/7, okay?

Third, I’ve done an experiment over the last week. I’ve run my own workflows through both APIs. I don’t care. Our ability to predict what we thought 2 weeks ago, when there was executive turmoil and everyone was using the Anthropic API because it’s just better, is limited. Claude, when we started the show, I was the only guy that used Claude. You guys probably thought Claude was ridiculous: Why is this weird guy using something that’s less than 1% of the market? Now everyone’s using Claude, right?

I don’t know what 90 days is going to bring, but everything could change in our development environments in a week, and OpenAI could win in agents, in autonomous agents. We’re just starting. Agents are just getting autonomous. All these goofballs on X with their Mac Minis and Mac macros, and 11 of them stacked in their closet—look, what have they shipped? Nothing, okay? It’s performative art. But autonomous agents are what we all are going to live with. We built some, and they could win. I mean, literally, they’re shipping it today. It could be the best thing in autonomous agents. Claude has a few now too—I mean, Anthropic—but that war’s just started. It’s just started. So I can’t honestly predict 30 days out.

So, does a trillion feel like Anthropic? Did that feel like a good deal on our last show? Am I sure it will 3 shows from now? I don’t know. Maybe they’ll invert.

Rory O'Driscoll

And if Jason is correct, look, I believe both of you are correct. You’re correct in the sense that AI’s the biggest story out there, and Anthropic has, in the last 6–9 months, been credited as the new winner in that space. So by definition, you’re going to have huge FOMO for that stock. You’re accurately representing the facts. Whether or not that will prove to have an overall holding-period return from here that outpaces the S&P 500, adjusted for the risk, is something entirely different.

Jason Lemkin

It’s hard to do in venture.

Rory O'Driscoll

The trajectory from here is actually pretty straightforward, which is: When you’ve got this level of white heat, you go public. Up until now, as I think I said in a tweet recently, we’ve had an entire life cycle of a technology explosion with very little access to it in the public markets, right? It’s hard to imagine going through the entire cycle of a boom and bust without letting the retail investors lose their money as part of the show. So that’s clearly the part that’s coming up.

My assumption is Anthropic will go public, and it will be a fabulously successful IPO, which is different than saying that at a trillion dollars you’ll feel regret about holding it 6 months later. It can be an amazing company, it cannot be worth a trillion dollars, and it can trade at a trillion dollars for a period of time. All those things can be true.

Harry Stebbings

What price does it go out at, Rory, if you were to make a guess?

Rory O'Driscoll

Your job as an investor is to tell me what something’s going to be worth 4 years from now, right? Your job as a banker is to tell me what it’s worth today. Those are different skills. One is about projecting the future; the other is about valuing the now. I’m not going to try and project the future on this now. My guess is I’d come in with a lower number.

But Jason is right about the present. Today, if it went public today, you would be fighting them off with a stick at a trillion dollars. You would have a fabulously successful IPO. And because those guys are wildly smart—there was a really nice profile of the CFO recently—and because the other thing that’s happened is the big counterattack on behalf of OpenAI has been: compute is the ballgame. These guys don’t have enough compute.

My guess is the Anthropic guys are saying, “Hey, we’ve got the better software. How do we get as much compute as possible? It takes money. This is America. Jensen will take an order provided we give him cash,” and therefore go public. So my assumption is these guys are going to go public, as we said, in the October/Q4 time period if they can, as soon as is humanly possible. Which is, of course, why they’re not taking the $800 billion valuation. I mean, why even pause for breath? Especially if it gives any kind of rights, which actually, at this point, I doubt it does. They are, I assume, heads down preparing to go public. Because as long as this mood lasts, you will probably raise money in the public market for that stock at a price you mightn’t see 3 years from now with 3 years of execution. They should go.

Jason Lemkin

After the IPO, what are they going to do? Are they going to do massive secondaries? Are they going to do PIPEs? Are they going to do weird debt-hybrid, equity-preferred deals? Because this isn’t the last time they’re going to need epic amounts of capital, right?

We’re talking about an IPO like it’s an event, which of course it is, but the ultimate—I think the meta question is: Where is it easier to raise ongoing capital, like perpetual capital? Is it really easier public than private for these guys? Not traditionally; the answer is, of course, but for these guys, is it?

Rory O'Driscoll

Yes. I think I’m going to say again, being opinionated on this one: yes, I think it will be, even though so far it hasn’t been. So far, the most stunning fact is that the biggest IPO in history is talking about raising $75 billion and the biggest private round in history raised $122 billion. What that says is that the biggest private round is bigger than the biggest IPO. That’s bizarre. We’ve kind of grown accustomed to it to the point where we don’t even focus on that fact, but it is, of course, absurd.

Given that logic, you would say, “Oh, stay private. You can raise more.” But I absolutely don’t buy that, as you know I’ve said. Yes, Anthropic could get 1 more round done, but the truth is the big liquid markets are public. I think they go public. You raise a big slug of equity out of the gate. You raise more. You obviously expand the capital base over time.

And then you have access to a whole bunch of other things. You have access to convertible preferreds. You probably can do more compelling debt structures. You become a viable player for more people to lend against. So if your business model is, as they think—perhaps insanely, but whatever—“I need to raise $200 billion, $300 billion,” and at some point you have to go public, why don’t you go public when the FOMO is at its peak, when your relative desirability is at a peak? It would be hard for me to imagine a better moment to go public if you’re Anthropic than this minute.

Jason Lemkin

But I’ll give you a counterargument, which I might be wrong on—admittedly, I might be wrong on—which is Figma and the need for capital. If Anthropic does not need any more capital, go public tomorrow, to your point, right? Or Q4, when you’re ready. Go public. Go public at a trillion. Trade up like Figma did to $2 trillion and ride it out.

But Figma’s down 87% from its peak. Imagine, for some reason, something like this happens to Anthropic. It still grows. I mean, Figma’s still growing at top-5% rates, right? But the market has fallen out of love with Figma. What happens if Anthropic IPOs at the perfect time for a company that does not need capital, but the markets fall out of love with it?

They get worried, like the construction costs go up. Those data centers we’re arguing over don’t work as well. Rory was right about the data centers, and it falls 87%. What happens to capital raising?

Rory O'Driscoll

Yeah, okay, respectfully, no. If everything that you articulated happens, then the person who’s public wins and the people who are still private are existentially screwed. The Figma soundbite, which I hate, of the 83% decline is more accurately represented as follows: Figma priced at $35 with sensible people. Idiots drove it up to north of $100 a share, then the same idiots sold it down.

So that’s 1 factor. But then, at the same time, Figma was at the tail end of a 20-year cloud-software boom, where the zeitgeist switched to the AI boom and the world fell out of love. So that took it from $35 down to where it is now, which is $20. That explains Figma.

If there is a world where the AI zeitgeist goes out in the next year or two, I think we’re at the start—maybe not the start—but we’re not going to hit a stage in the next year or two where people are saying, “Oh my God, the story’s dead. We’ve moved on to the next thing.” You’re closer to the start than the finish of the zeitgeist.

I think there could be a hit to the stock, and I’ll come back to that in a second, but I don’t think it’s the same as Figma, where literally it’s like everyone saying, “Your operational results are amazing, but we’re in love with someone else, so we’re just not even going to talk to you.”

If Anthropic were to go out and if the stock were to go down, it would be because, oh my God, it’s overvalued, and this revolution’s going to take 20 years, not 5, so maybe overpriced. And in that case, you’ll sit there—and I’ve been in much smaller scale on this—and go, “That’s really bad. Our stock is down. On the other hand, we’ve got $50 billion in cash, and our head-to-head competitor’s still losing money in private. We win.”

So, no, there’s no argument for overstaying your welcome in the private markets at this point. None.

5. Claude Design Threatens Figma

Harry Stebbings

Jason, you brought up Figma, and you brought them into the conversation. Tying that to what we were talking about, Anthropic launches Claude Design.

For those who don't know, it's their competitor product to Figma, very bluntly. Obviously, Figma as a result was hit, and its price was hit significantly, as was Adobe's. Jason, you actually did a like-for-like comparison, and you've tried Claude Design extensively. I'd love to hear your thoughts. What were your reflections? Does it compete with Figma in a very meaningful way, and how do you leave feeling?

Jason Lemkin

I leave with a lot of anxiety over Figma and others, but not for the reason the yahoos on Twitter said. There are 2 things to know about Claude Design if you haven't used it. One, it is definitely better than the design tools that were in Claude the day before. You've always been able to design a website that looks exactly like every other website built with Claude Artifacts and purple gradients, and they all look the same. Half of Demo Day looks like it was built in Claude Artifacts, at most. You can smell them in 60 seconds.

Design was always there. It just wasn't what a lot of designers would call design, but you could always design a website. Now Anthropic did something important. We'll see where it goes. They didn't just improve it; they built an application. Claude Design is an application.

There are only so many Anthropic and OpenAI applications. There are skills, workflows, prompts, and little things you can do. They went to the trouble to build a design application that works. Whether it's 50% better or 200% better than the day before, you can design better websites and better properties in Claude than you could before. Does that mean you can build what you can build in Figma, let alone Illustrator? No.

A lot of the yahoos said on Twitter, "Of course this isn't a threat to perfect design, to taste," because we're all about taste now in AI. Taste. We have no threats because of taste, and the combination of taste and moats means we're unassailable. So, your typical designer, hoping to afford tickets to Coachella, who takes 2 weeks to respond to a ticket to develop an asset—are they going to switch? Maybe not. They're not going to switch from these things.

But it means normal people can design stuff and get into production much faster. So I think it is an existential threat. It will maim and nibble at Figma more and more because if the 3 of us wanted to build an app together, and we don't want to wait for a designer to turn it around in 30 days and give us a Figma file or a still, what the hell do I even do with a Figma file? I can stick it in Replit. Now I can just do it myself. So we will bypass designers more and more. That's the risk.

I do not believe Anthropic will ever build a direct Figma, Illustrator, or Adobe competitor. They don't have to in order to maim them. So, yeah, is it a design tool? Maybe it's design-to-production, like some folks call it. Who cares? If it maims you, it maims you. I think people are missing the point.

The meta question is that I saw no one talk about it: it is an application. It comes up as a full application. It has sharing. It has users. It has hierarchy. It can save assets. I do not believe there are many other applications that Anthropic or OpenAI have built.

If you look at what the public markets are scared about, they're scared about a lot of things, but we make fun of vibe coding. "Oh, we're going to vibe-code our Salesforce." But what if they start building A-tier applications, not just prompts and not just outputs? It's something to reflect on.

I'll start to ramble, but I'll give you one last story to compare. The oldest piece of software we use is Marketo. It's a terrible email marketing service. About 2 weeks ago, it started to violate the CAN-SPAM Act. I started to see things on Twitter: "Jason, how come I can't unsubscribe to your goddamn SaaStr newsletter?" They pinged me again and again. I don't know. Then you get more of them, right? That's when a bug has been introduced into the system.

We flagged it for Adobe and Marketo. They said it was unfixable. It was a CAN-SPAM violation. This was a week ago. Then they said they would only help us if we got on the phone with their engineering team yesterday. We did. They did nothing but blame Salesforce and said they could not commit to a fix.

So my point is, this is what old software looks like. If Anthropic and OpenAI are going to build applications, sell them all as a bucket. Keep the gems, don't get me wrong, but they're coming for old software. They're coming for old software. This is an application. This is not a prompt, but it is not going to maim Figma next quarter. Those can both be true: we won't see it in Figma's numbers next quarter, and over 4, 6, or 8 quarters, it will maim its growth. We won't want to use grandpa's software anymore.

Rory O'Driscoll

Honestly, I want to ask a lot of questions because, one, that's what we do here. Two, you've used both products, and I tried to look at some demos today, but I have a ton of questions. It's like the Emerson quote: "If you understand one thing, one man well, you understand every man. If you understand one competitive market well, you have a kind of a framework for all these markets."

I just want to drill down on this because it's the same question in every market. A couple of things. One is, how do you think it impacts—I mean, the Canva design engine is at the heart of what they've done on Claude. How do you think this impacts Canva versus Figma?

Jason Lemkin

First of all, kudos to Canva. If you use Anthropic—I mean, Claude Design—it exports to Canva. So you can import from Figma and export to Canva, should you choose to.

But what's happening is product teams and engineering teams are already changing. If we go back 12 months ago, you had design. Back in the day, when I worked at Adobe, there was a design group. We weren't allowed to design anything that was public-facing. You'd file a ticket, and 60 or 90 days later, you'd get a PDF of what your website had to look like. Then your product team would have to figure out what to do with that PDF, argue with your engineering team, and months later you could ship it. That's the way software used to work until not too long ago.

Then, as everyone started working in Claude Code toward the end of last year, product and engineering teams just started working in the code together. Product teams would first vibe stuff in Replit and Lovable on their own, right? Now a bigger and bigger deal is that they're committing to the codebase. Product teams are able to do this.

These PRD teams are becoming more cohesive, and design is still way out over there in its own hipster land. As these combine, everyone's going to want to work directly in Claude, Codex, and Claude Code. It's not going to be that a designer enforces a collaborative hierarchy on Figma. The designers held the product organization hostage before AI. They hold them hostage. It's the worst.

Rory O'Driscoll

It's worth just making a high-level comment here. When you're saying "design," implicitly what you're saying here is that it's the Figma digital design, websites, and apps world—the software part of design—not the part of Canva that is printing out real-world pictures and posters, and the physical design where it's decoupled from software.

Jason Lemkin

You're right. Design's such a confusing term, right? It's a bigger threat in the short term to things like Gamma. It already makes slides like Gamma. It's not a threat to Canva today, but every single thing you do in Claude Design that you don't do in Canva, Figma, or Gamma is a threat to them, even if it doesn't kill them. Every single thing you do is.

Harry Stebbings

I think the smartest thing Jason always says, Rory, which always sticks with me, is just the element of maiming, which is quite hard to deny. If it takes 20% to 30% away because you're already there and it's easy, that's very meaningful.

Rory O'Driscoll

It is meaningful. And again, I'm not diminishing that. I always thought one of Google's interesting strategies for a decade and a half was their strategy with G Suite. It was email, effectively Docs and a spreadsheet, which just gnawed away at the bottom end of the Microsoft Office suite.

If you fast-forward a decade and a half, you get both sides of it, though. Basically, they could spend $1 billion yanking Microsoft's chain on a $40 billion business. It's exactly what Jason says: "I'm messing with your head, I'm maiming you, and it's not core to me, so I don't have to win here; you just have to lose."

On the other hand, 15 years later, Office is still a plus-or-minus $40 billion business with a lower growth rate. So there is a range of outcomes: it gnaws away at the low end, takes away a lot of users but only 5% of the revenue, or it bites 30% of the users and 25% of the revenue, and then it really impacts. Those are big differences in terms of value creation.

Jason Lemkin

We'll have a maim meter in board meetings: "How much have we been maimed this quarter?" It'll go from 1% all the way to 50%. Our agents will decide. We don't let the founders decide because they're always at 1%, or they're the overreactive one at 50%.

Rory O'Driscoll

This is a cue for the Monty Python “It's Just a Flesh Wound” sketch.

Jason Lemkin

It is. It is like that. That's what you're going to say: "It's just a flesh wound. I lost my arm and my leg this week to Claude."

Rory O'Driscoll

There are 2 things you said that resonate. One is, anytime someone can give away something at the margin for free that takes away some of your low end, there's an impact to that. Bundling is a bitch, and they're effectively bundling here.

The other thing that I think is more important is what you said, and I hadn't internalized this: if the design flow is a preamble to a technology-build workflow, and the technology-build part of that workflow is already automated using code—Claude Code—then you're right about the incentives.

Where you get scary as a standalone company is if the other guys have a better-together story. And you're right: to the extent that design, product, and engineering can all be in the same tool, there's probably a whole bunch of embedded efficiencies there.

OpenAI did a ton of these. Now, if you look back on the GPT Store and GPT plugins—it's just fun to do it now—every single time Twitter gets all excited, “This is the end of everything. XYZ company is screwed,” and it turns out, 2 years later, no one even remembers that, and the thing's been deprecated.

But you're right. In this case, if you think of design not as a standalone category like Canva, but as the front-end window into product and engineering, and if you think that software is the mother lode, and software coding is the mother lode of enterprise adoption, then you probably get enough effort behind it to build a credible product. That's what you're saying.

Put bluntly, at some point, even Anthropic has to start allocating resources at the margin. Implicitly, you're saying there'll be a big enough team here to build, as you say, all the app features to make it a viable competitor.

Jason Lemkin

Yeah, and not only does it export to Canva, it exports to Claude Code. They're integrated. So if we're moving quickly, we're not going to wait.

Listen, I still want my human designer to deliver my amazing homepage of my app, my amazing splash screen, my amazing assets, but we're shipping features every day, guys. I don't have time to wait anymore. We're going to do it in design. It goes straight into Claude Code, which we already run our company on, we already ship on, right? It's fully integrated.

If the humans have time to redesign it later and make it better, great. No one's saying it's pixel-perfect. Don't have time. I don't have time.

So the only reason I think they won't entirely abandon it is it's part of the core. It's part of the Claude Code core, and it is something you can already do in Claude Code, or even in Claude, just crappily. So they're improving. The closer it is to the core, the higher the chance they'll maintain it, and it's not a dalliance, right?

Not only are the designers stressed, not only are Dylan and Mike Kenna at Brick stressed, they're stressed at Lovable, Replit, Vercel, and Gamma too, because the rate of competition is unlike anything we've ever seen before. And I will say, one thing I know from all these founders: they're brutally aware of it.

The older CEOs are hiding from it. “Oh, our next agent will catch up. Oh, we'll catch up later in the year. Oh, you haven't seen our next...” This is what I hear from pre-AI founders: “We'll catch up in the next release,” they say so confidently, right, as they go off to complete their triathlon.

The AI-native CEOs are frigging all over this, okay? In 60 seconds, you get a Slack back. They actually know what's coming a month later and what all their competitors are launching. They're already all over it.

I hate the 4D chess game, but if you don't play 4D chess, if you're at the core of AI, you're going to lose. So they're all playing this game. It just gets harder every week.

6. AI Funds Chase Growth

Harry Stebbings

Accel adds a $4 billion Leaders Fund to follow into hot AI growth rounds. Sequoia, 4 days ago, raises a $7 billion growth fund. Is this just further compounding what we said, which is that this is the game today, and the leaders see it?

Rory O'Driscoll

It's definitely the game today. It's definitely what the leaders see. Obviously, the question is whether it's correct. But yes.

Harry Stebbings

Do you think it's correct?

Rory O'Driscoll

Directionally, yes. People are saying companies are staying private longer and outcomes are bigger. So capitalism is doing what capitalism should do. It's raising money to put into these companies.

The thing about growth is this: when growth is sexy and attractive from a fundraising perspective, it tends to be hard to make money at. And the way growth makes money is either overall valuations are down, like in 2022, or there's some insight that the growth investor has that the wider market hasn't figured out yet, like ChatGPT really matters and you should buy OpenAI.

So both of those conditions were true in 2022 and 2023, which is why those funds are going to be awesome. If you have a situation where capital is plentiful for those rounds and everyone understands AI is amazing, then, by definition, you've eroded the 2 things that gave you excess returns.

I'm not saying you still won't make great returns. It won't be as clear or as compelling as it was when it was unpopular.

Jason Lemkin

The other thing is there's room for the capital. And what I mean is, if you look back at 20VC Fund 1, a $10 million fund—I'm speaking for history for Harry—back then, the strategy was to get into hot seed or A rounds where he could write a $50K check, $100K check. There was always room. There was always room.

To some extent, it's still true today; it's just those hotter checks are at much higher valuations. Fast-forward to today. If you're a relationship builder, if you drop by and meet with Dario, if you schmooze with Sam, there's room in the round. You don't get cut out.

Don't get me wrong, people are getting cut out of these rounds, but if you're Sequoia or Accel and you're a good schmoozer and a people person, and you show up to poker night and you do all the right things, most of these rounds you're going to get an allocation. You may be the 4th name on the press release, but you're going to get an allocation.

And so, in a way, the fund sizes don't even sound that large. If you get 20 or 30 checks, they're not even that big.

Rory O'Driscoll

And that's totally fair. You're exactly right, Jason. When you're raising $122 billion, it turns out there's room for everybody. Even when you're raising a miserly $30 billion in the last Anthropic run, you're right, there's room for everybody.

Jason Lemkin

Well, people get cut out. They're desperate to get into Anthropic, but if you have the relationship and you really build it over a period of time, when they look at the spreadsheet for the round, they're going to put you above the fold because, “We like Rory. He's great on the 20 pod. He came by the office. He loves us. Fine, we'll give $20 million to Rory.”

Do you disagree, Harry?

Harry Stebbings

No, he doesn't, because his first sentence—his comment was, “If you're going back to Harry's LP who wants Anthropic and nothing else...” What I like about Sequoia is they know how to make money, right? They're like, “Let me get this straight. I have a good relationship with these mega-companies. I'm a name that they'll want even at this stage, and I got a bunch of LPs saying to me, ‘The only thing I want is this,’ and I'm in the middle. Hmm, let me think about this. I should take their money and give it to these guys, charge my percentage, and call it a day.”

As long as that works, the marquee names who can do that are going to do it.

Some of the package pricing on these SPVs for Anthropic has been some of the most egregious face-ripping I've ever seen. I mean, 8% up front—crazy.

7. Rippling Resets SaaS Expectations

One that I thought was fascinating was Rippling crossing $1 billion, growing 78% year on year. Is this the new bar for a great B2B IPO?

Rory O'Driscoll

Let's just step back even 1 level beyond the IPO. I think what this sentence, if true—and you never know, but I believe it to be—exposes is that the whole “SaaS is dead” meme is bullshit. Low-growth SaaS is bad, and high-growth SaaS is good.

This is a market where we can talk about the reasons why. If you're doing $1 billion, growing at 78%, then it collapses the entire discussion about “SaaS is going away,” “SaaS is bad,” and “What's it worth?” What it really points out is the real objection to these public SaaS companies is not, “Oh my God, you're SaaS”; it's that your market has now topped out, and your growth rate is 10%.

If Rippling is growing at that rate, that's amazing, it's compelling, and they'll get an excellent IPO.

Jason Lemkin

And they're accelerating. The crazy thing is they're accelerating. So if they were at less than $500 million 11 months ago, which I do know, right, and they're at $1 billion, growing 70-something percent, do the math with me, Rory. They're accelerating. It's not just 70%, which is enough.

And you can say that's great for SaaS, and it is, right? As Parker said to me, “Not bad for a SaaS company,” right? Kudos. And we can talk about what that means. I'm not even sure what it means, but not that many are accelerating like this.

This is not good news for anybody not accelerating at scale. It's not even 70%. It's accelerating from under 50% to 70% in a year.

I mean, he's a candidate for CEO of the year. Not to disparage Deel. To drive that level of acceleration, I don't even know how to do that. Unless it's people buying tokens, I don't even know how to do that in today's world.

Harry Stebbings

I mean, a bold candidate for CEO of the year. Respectfully, he's got a bit of competition, Jason.

Jason Lemkin

Yeah, but they've just launched their agent. Driving that level of acceleration without a massive AI tailwind just blowing your phones up, right? Bring them on. We'll talk about it. I don't know how to do that without AI, right?

Rory O'Driscoll

I think it's what they refer to in sporting terms as a win against the run of play.

Against the odds, you’re playing the SaaS game where everyone’s walking around saying the world is dead. Yes, Anthropic’s putting up 600–700% growth, but you put 78% growth and acceleration up in a category where most ill-informed people were saying you can’t do that. Maybe the better, clearer expression is highest outperformance relative to, quote-unquote, “market expectations.” I think that’s a clear example of it.

Jason Lemkin

Yeah. It’s better than Figma by the numbers, right? It’s better than Figma by the numbers—by far, by far.

Rory O'Driscoll

And what it just says—serious comment—is there are markets where the entire agentic AI vibe-coding discussion is rubbish, and one of them is financially related stuff and payroll. I’ve run a business, and you have—you guys both run businesses. You can screw up on a lot of things. You screw up on employee payroll and you pay them at 3:00, they’ll be in your office at 3:01. You can’t get this wrong.

You’re not interested in vibe coding. You’re interested in having it right. You’ve got legal and statutory obligations that carry criminal penalties if you don’t pay your taxes. You’re like, “I want to outsource this to someone wildly competent, have them take the responsibility, and no, I don’t want nondeterministic processes, you moron.” These are entire businesses that might have some impact at the margin in terms of agentic efficiency, but this core business will be there in 5–10 years’ time.

We’re actually talking about this a lot internally: what gets eaten by AI? What doesn’t get eaten by AI? What’s defensible? Payroll is one of the best examples of this. It’s just something you do. You might build software better using AI, but the core value proposition is just orthogonal to AI, and it has to be done right. So, big category. It’ll go public. It’ll be a great outcome. Good for him.

Jason Lemkin

I think the same with a lot of the fintech players that we see today. Your Ramp, your Stripe, or your Air Wallets of the world are not immune.

You know what, though? I know it’s not to harp on the M&A episode. The moat is stronger and it is less impacted by AI, but everything I view now is how well it works with our agents, how well its API and workflow works with our agents.

For example, after SaaStr Annual in May, we’re going to build our own AI VP of Finance, and the number one thing we’re going to do is automate collections. We’ve been on Brex for 6 years. You know what the first thing we’re going to look at? Which API works best with our agents.

I don’t care what Ramp’s dashboard looks like. I don’t care. I care how our agents work with its API, and we’re going to pick the best one. That is just starting, but it is a BFD. It also means that folks that seem to have a huge moat may see it weaken.

I’m not just being a yahoo on X. We’re going to build an AI VP of Finance, and it doesn’t care what the UX is. It doesn’t care.

Rory O'Driscoll

First of all, you’re right. You’re absolutely right, and therefore the payroll and AP and collections provider that has the best API will win. But I’ll tell you something else: you’re not going to build an entire fintech stack. So someone will get those dollars. You’re right.

Jason Lemkin

Yeah, someone will, but it could be a new vendor. It could be a different vendor.

Rory O'Driscoll

You’re right. I mean, this gets to the heart of it. There are 3 or 4 vendors of AP-related stuff. You’ve got Ramp, you’ve got Bill, where we’re involved, and you’ve got Brex, whatever, right?

Whichever of those doesn’t have an API-forward product will lose market share. Whichever one does will gain market share. If all of them are dumb enough not to do it, then a new vendor who says, “I’m an API-first product,” will get your business and all the people like you.

I think you’re correct in that. A platform shift, which is what we’re dealing with at this level, has implications for everyone in the tech stack. But if you are doing something like what Ramp, Brex, Bill, and all these people are doing, the core thing you do itself won’t be replaced in the way, for example, Figma might be. My point is there are degrees of change here.

Jason Lemkin

But the degree of comfort that we’re protected—the degree of comfort that we’re protected—Ramp has asked us to switch from Brex for 7 years. I bet we finally switch over the summer, and it’s only because of the agent.

If it wins the bake-off, we will switch in 1 week. We will switch, and we will never go back to Brex. Same with Marketo and HubSpot, by the way. We’re leaving Marketo because of this drama, their crappy API, and their canned spam. We will leave Marketo this summer for whoever has the best API, and it probably won’t be HubSpot.

8. Salesforce Builds Agent Fabric

Rory O'Driscoll

Which is why we should pause here, to your point. Salesforce just announced an entire headless API strategy. What are your thoughts?

Jason Lemkin

While we debate where to move our Marketo data, we’re already using headless Salesforce to move all of our Marketo data agentically over to Salesforce. It’ll be done in a couple of days.

Rory O'Driscoll

See, that’s a great story because—give him credit, give Marc Benioff credit—he’s like, “If you can’t beat them, join them,” as they say in private, right?

Remember, it’s less than a year ago that there was talk about, “We’re not going to let you have your data,” and now we’ve gone to getting a headless one out the door. You’re an example of where he’s taken share from adjacent companies like Marketo just because you’re doing that, and that’s your point.

Jason Lemkin

The only thing I will say is, listen, we’re already living the headless vision. We don’t log into Salesforce. It’s our hub. But a lot of this stuff is going to run on MuleSoft in a couple of months or 6 months.

This is also, with love, a classic B2B. Anthropic drops a design tool, and we can use it in an hour. This is something that will be dribbled out in classic B2B fashion. It’s just different worlds.

Harry Stebbings

Can we provide some context also for those who haven’t heard, in terms of Benioff’s move with headless that he announced? Reset the scene there, because I want to ask some questions.

Rory O'Driscoll

Sure. Traditionally, the Salesforce app really had 2 components of value. There was the user interface that every sales rep, or everyone in the organization, uses to input and output information and effectively record the work they’re doing. At the back end, you have this massive database and workflow that records all the information and allows you to track customers, leads, pipeline, and all that relevant stuff.

So there are 2 components of value, and by offering a headless offering, what he’s basically saying is, if those people who are doing the work are replaced by agents doing the work, then they don’t need my UI anymore, right? Because the people aren’t there anymore. They need a totally different agent-based UI.

But what I’m going to do as the leader of Salesforce is still allow them to access the database side of my product, which means I keep my value even in a world where most of the selling or customer support is not done by humans typing into the Salesforce UI, but is done by agents proactively going against the Salesforce back end.

So he’s given up trying to drive per-seat pricing to preserve long-term value, because I think he correctly identified that the real long-term value Salesforce has is that it’s taken us 10 or 15 years to get all those integrations in place and all that data in place.

If you make it easy for Jason’s agents to work with Salesforce, then he might not get around to killing you for the longest time. So that’s the big move they made in the last few weeks.

Jason Lemkin

I think, like Claude Design, though, everyone completely misunderstood what headless Salesforce is. Salesforce is already headless. This is actually pretty crazy. Mark and Parker, to their credit—JFC—in 2006, they launched an enterprise API when this was seen as impossible.

You could not build or allow third parties to integrate into enterprise software. It was too risky and too problematic. They opened this platform up, and they have 20 years of it getting better.

I will tell you, for all the APIs that our agents use—our AI VP of Marketing and VP of Customer Success—Salesforce is the best API out there. It is the best API out there. It crushes everybody. This includes all the new guys, everybody.

It’s because they built this for 20 years, and because the APIs are so good, the agents can work with it. It’s not a problem. What headless is—and this is what Marc’s out there positioning, I mean, the greatest marketer in B2B, right?—it’s already done this since agents started.

What he’s really pitching, which is the bigger threat to these leaders and the bigger opportunity, is an agent fabric. It’s the layer that manages all your agents. If you really look at what this is, it’s about Salesforce saying, “We are going to be the fabric to manage all of your agents—your 20 agents, your 50 agents, your 100 agents.”

Some of them will be built on Salesforce. A lot of them are going to be built on MuleSoft, as crazy as it sounds—their platform, which has been renamed. It will be in their data platform, but we will be the layer to provide the context, the guardrails, the management, the security, everything.

The biggest issue for 2027 is agent fabric.

Jason Lemkin

People are under-discussing this. They're all talking about evals and this crap. This is what really matters in the enterprise: agent fabric. You can't let these crazy agents run amok.

Harry Stebbings

I'm sorry, I'm just going back. When you say agent fabric, you're saying agent orchestration, management—

Jason Lemkin

It manages everything: all the governance, all the security. It knows—here's the key part—it knows what every single agent is doing in real time. That's the fabric. That is more than orchestration, okay? It literally knows every piece of data, every operation, everything that's happening through 200 agents running 24/7 in parallel, with multiple sub-agents in them.

Who's going to go to poor Jacob O'Driscoll, the CIO of wherever, and say, “If there's any security breach, you lose your job”? It's his worst job. He wants a trusted agent fabric to manage these crazy agents his team is deploying. They can't be done out of ChatGPT. They can't be done out of Base44. Geez, JFC, this is a security nightmare, right?

Even this week—and I don't want to get into it—two leading vibe-code platforms arguably had massive security issues. I don't want to talk about them, but this is only going to compound. As Mythos comes out and finds every security breach in nanoseconds, I need an agent fabric I can trust, not someone that came out of YC and claims they have an orchestration platform.

But I don't know if Salesforce can deliver it because it's so effing complicated. But if you're in their ecosystem, if you commit to everything—when most folks hear about Salesforce, they think it's CRM. It's 14% of their revenue, okay? If you commit to everything—e-commerce, marketing, data analytics, Slack—and you run your whole business on Salesforce, this is the old SAP pitch: “We will give you the agent fabric so you can accomplish everything you want.” This is an agentic platform you can trust. This is the big bet coming, and this is all the warm-up phase for agents. Enterprises need an agent fabric they can trust.

Rory O'Driscoll

Agreed. Because you're empowering these software agents to do things, to change things in your systems, to upgrade things, to approve orders, to make commitments. How do you audit what's going on? How do you know? How do you keep control of it? That's—you’re exactly right. That's going to be the issue.

Jason Lemkin

I've got 300 agents doing accounts receivable. I've got 200 updating our documents. I've got all of these interacting: our autonomous customer-success agents, our economic data analytics agents. Who's going to manage all this?

Orchestration is the nerdy term, but most folks talk about orchestration as just a limited dashboard on top of a couple of easy APIs to connect with. It's fine for a startup or for a team that has human resources, but how's an ordinary company going to manage these agents? How the hell is an ordinary company without a team of agent-deployment experts going to manage these agents? They'll go rogue if you don't manage them.

Rory O'Driscoll

This analogy may be totally useless, and it's only because I've been doing it a long time, but I remember in the late '90s, when online commerce took off and people really started getting a meaningful percentage of their revenue from online commerce, you had all these executives who were retailers to the core. What do retail executives do when they want to know what's going on? They go walk the floor. They want to know what's going on in the shop.

Even if you're running a 5,500-store chain like Walmart, Sam Walton used to walk around, touch the merchandise, see what's going on. Suddenly, the whole thing's going on online, and you just don't know. People are clicking. That's all you've got. You saw a whole wave of companies doing analytics around how to track your website, because the big guy just wants to know what's going on.

That was the value proposition. We did NetGenesis, we did Omniture, and made a lot of money in that space. The value proposition at its core was, you've moved to this new way of doing business. Senior people want to know what's going on. Frankly, AI is way more powerful than that, because at least in that case you had very deterministic—“I'm selling stuff at a certain price for a certain thing”—and even then you wanted to know. You didn't want it dumb.

In this case, you've empowered your agents to make decisions. Now you're the executive in 2026. You're going to know what's going on. Because I think about it, even what's Gong? Gong is all about listening in to calls to understand what people are saying. Once you have AI agents doing all this stuff, you're going to want to know what the AI agents are doing. I think, Jason, you're exactly right. This is going to be the huge thing. How do I think about what my little automated bots are doing in my business?

Harry Stebbings

My question was, is Salesforce best placed to be the agent fabric, or is someone else better placed? If that's the whole holy grail—

Jason Lemkin

Of course they're well-placed. It's a bad analogy, but just like in the end Google, Microsoft, et al. were well-placed for the last generation of AI, if they can get their rears together, the leaders are well-placed. Salesforce, Shopify, Datadog, Databricks—if they can execute faster than they've executed the last 20 years or 8 years, however old they are—of course CIOs want to buy from Salesforce. Of course they do.

But it's much bigger than buying one Agentforce agent. They want this whole agentic fabric. Here's my meta-point to folks: it's an opportunity where you have time; it's just not infinite. If you're not building up that whole fabric, right? That's why I'm a fan of Marc. This may not get there. It is a bigger vision than it looked. It's not just headless, like all the yahoos said, okay?

This is a complicated vision, and maybe they won't get there in time, but at least he's driving the right vision and forcing thousands and thousands of people to deliver against it, right? I'm much more worried about folks for whom it's more performative, right? You've got to work as hard or harder than Marc to do it.

But I don't mean to be repetitive, but the saddest thing is when you have an install base and you're not delivering the agentic solutions they want. This is the tragedy of 2026 and 2027. I have the customers, but Lagora or Replit or whoever—pick any application you want. What's the AEO one you invested in, Harry? What's it called?

Harry Stebbings

Oh, Peak.

Jason Lemkin

Yeah. HubSpot launches its thing and it's a dud. It should've been Peec or the other one. It's just a tragedy because HubSpot has 280,000 customers. It's a tragedy you did not deliver them the best-in-class AEO agent. It's a tragedy—it's not just a test or a miss.

So that's why I think Salesforce is extremely well-positioned, and we are right to be stressed, like everyone is stressed, that they will achieve it on time. We're right to be stressed.

Harry Stebbings

It's really interesting. They've had Inbound be bought by so many people. Elad Gil tweeted a load of predictions, and one of his predictions was that a lot of these AI companies should actively sell and try to sell. Do you guys agree?

Jason Lemkin

To whom?

Harry Stebbings

To HubSpot. To sales tech now.

Jason Lemkin

Literally, Harry, I was on the phone last week with the CEO of a $20 billion or $30 billion market-cap public company, okay? Doing massive revenue. He's like, “I get these M&A...” We were talking about M&A a little bit. I brought it up; he didn't bring it up. I'm like, “Well, go buy some of these kids, right? You have the base, right?”

He's like, “Well, everyone wants a billion on $5 million in revenue after their last round.” He's like, “It's almost a waste of our... I have a corp dev team, but I haven't seen a single one that I'd want to buy that will sell at a valuation that makes sense.” So Google can buy them, but HubSpot—what's HubSpot's valuation as we record this? In the teens of billions? They just can't afford $1 billion for every YC startup. They don't have the money.

Rory O'Driscoll

Yeah. I would say try harder; stay close. There are going to be plenty.

Jason Lemkin

Plenty of exits? How many whizzes can Google— Is Google really going to buy, though, Rory?

Rory O'Driscoll

No, I'm not talking about those kinds of exits. I think, to your point—what's that?

Harry Stebbings

$12 billion—HubSpot.

Jason Lemkin

So they could afford $50 million to take a risk, right? Not a billion.

Rory O'Driscoll

I mean, it's a fun fact. You know I like this space, but someone told me that—is it G2 Crowd?—said there's, like, 250 AEO/GEO competitors out there. Maybe you can't afford numbers 1 to 5, but somewhere between 10 and 250, there's going to be one that has a good product. So I think Elad's entirely right.

Jason Lemkin

But isn't Elad saying sell at $1 billion? I think that's implicitly what he's saying.

Rory O'Driscoll

Yes, and if you can, great.

Jason Lemkin

Sell now for an easy billion, right?

Rory O'Driscoll

Yeah.

Jason Lemkin

I'm optimized for $100 billion outcomes at my fund, but sell for a billion now while you can. Honestly, that's what I think he's saying, right? I would agree. His other point was you should have an exit discussion every year with your portfolio companies, right? I'm going to have one tomorrow. It's a great tip. He's super smart.

I just wonder where the $1 billion-at-$5 million ARR exits are coming from. I want their number. Give me their WhatsApp or their text, because I'm going to send them a couple of deals right after we get off this show.

Rory O'Driscoll

And my point is this. In many respects, both of you are saying that, when you listen to what Jason says about the public markets and the dilemma they're going through, they absolutely should be looking to acquire some of these things to get some of this technology.

Rory O'Driscoll

And maybe what we’re really saying is the biggest rate-limiting factor is not their unwillingness to do it, but the prices at which the venture crowd think we’re gonna get for them have made it hard to make those transactions. It’s been my experience that if that’s the case, in the end, things true up.

Harry Stebbings

But you know what’s tough? I picked on this HubSpot AEO product, okay? I picked on it, but I didn’t know they bought that company for $30 million months ago. So everything just gets stale so quickly.

I’m sure HubSpot sat around and said, “Listen, we need to be in this place. It makes sense. We need to be in AEO. What can we afford? Well, Harry just funded this one. We can’t buy that for $30 million. There’s the other one in the US; we can’t afford them. What’s available that we can afford?”

And I think 5 years ago, that was a good strategy, right? Because the world moves slowly. Now you buy something that—let’s assume whatever company they bought was competitive 4 months ago—it’s just not competitive at today’s pace. I don’t mean to pick on them, but it’s why M&A is tough, right? Who wants to buy something that’s gonna get stale?

Look at poor TBN. We can’t even see it on our feed anymore. It’s disappeared since the acquisition, right? Gone. So why even buy any of these things if they instantly become stale?

You do these tuck-in acquisitions, which used to be—go back to corp dev—the best strategy was the barbell: just start to invest. Buy something small for $50 million to $80 million for product, a couple million in revenue to prove it works, and rebuild it over a year. Rebuild it natively on Salesforce or HubSpot, or go big, right? Because you’ve got scale. But it’s tougher today to find the gems that want to sell with product-market fit cheap.

Jason Lemkin

I don’t think you’re saying it’s tougher to find them. I think you’re saying it’s tougher to manage them and preserve the urgency and the speed.

Harry Stebbings

I mean, even OpenClaw, that dude’s just on the TED circuit now, right? OpenClaw may be obsolete in a couple more weeks.

Rory O'Driscoll

The 2 more for me are Snap and Cerebras.

Jason Lemkin

Yeah, I think you have to do Cerebras first. I just have a predilection for good over bad.

9. Cerebras Returns To Public Markets

Rory O'Driscoll

Okay, let’s do good over bad. Cerebras files for an IPO. It’s the second time. Some of the concerns that were brought up last time, in terms of a dependence on G42’s revenue and the revenue concentration they had, have been resolved. How do we feel about this? Is this gonna go out well? They’re now at—where are they?—$510 million in revenue in 2025, up 76% from $219 million in 2024. They’ve done a great job.

Rory O'Driscoll

Absolutely. That’s why I wanted to cover it. I think it’s a great classic venture deal. Credit to Benchmark, Eric. Credit to Steve at Foundation. Credit, obviously more than anything, to the team. You know, this is a 10-year journey. And I think, in a way that wasn’t true a year ago, they got the elements of success in place.

I mean, you look at the P&L, it’s a little noisier than at first glance, because you at one point said, Harry, “It’s profitable.” It’s not. It had some weird reversal of liabilities. At the operational level, it’s still losing money, but the big aha is they’ve effectively done exactly what it takes. They’ve proven the product.

And stepping back, this is a semiconductor company, potentially one of the very few startup semiconductor companies in the last decade and a half. The product they make is a big-ass wafer-scale chip that’s really good for inference because it’s really fast, and obviously it’s very optimized for AI. And if you read the founder letter, it’s been optimized for ASICs. They founded it in 2016.

What they’ve done really well is—this is a very hard step. This is a very hard market to enter, no matter how good your chip is, because there are only a small number of big buyers, right? Obviously, the hyperscalers, right? And some of them have their own chip; some of them might want to trust you.

These guys have done a couple of things to parlay their way in. They’ve done some big deals in the Middle East with folks who’ve been willing to use the product. They started to offer a cloud offering, and one of my companies has used it, whereby they are effectively saying, “Our chip is so good and so fast that we’ll offer inference services on a cloud basis, and when you use it, you’ll be like, ‘Oh my God, this is really fast.’”

And as I said, one of my companies has had that experience. It’s really fast, low latency. It’s a great product. And those things allowed it to prove that the chip worked, and in the last 3 months they’ve signed a deal with OpenAI and a deal with AWS, right?

And the OpenAI deal is for the usual $20 billion of commitments. Who the hell knows what that means. But the point is, they’ve gone from niche to mainstream, and they’ve clawed their way into the mix. So I just give them huge credit, because that’s a long journey. It’s a hard journey.

And a year ago, it was easy to sneer and say, “Yeah, you’ve got a bunch of investor contracts in the Middle East. It’s all bullshit.” Now you’ve got the inference business—you’ve got the service business. You’ve got the incipient contracts with 2 of the largest players in AI. It’s a credible play.

So I think it gets done, and it gets done well, and it deserves to get done well. And now I’m on my soapbox. This is just great. It is just great that venture does this kind of thing. A 10-year journey to finance a new chip for a new use case that’s complex technically, a complex business, and they pull it off. I hope they make a ton of money.

Jason Lemkin

I agree. I really like Andrew Feldman, too, the CEO. He’s a really good dude. Where does this go out at? Groq sold for $20 billion. Does the benchmarking of Groq enable this to be a $25 billion IPO?

Rory O'Driscoll

When things are valued on a P/E or an earnings basis, you can have an opinion, a meaningful opinion, on where they should trade. In this case, it’s gonna be so much narrative-based, it could go well above that.

I mean, again, we talked about this a while ago. If the leading player is worth $5 trillion, and then the next 2 leading players after that are—well, you’ve got AMD, and then you’ve got in-house silicon from Google and Amazon—this is the only other standalone play you can make, right? I mean, do the math here. 1% of NVIDIA is $50 billion.

If you just think of it as a call option on some percentage of a $5 trillion market, you can see a very big outcome here. Now, you can squint the other way and go, “Oh my God, they’re never gonna...” It’s a high-beta area. Again, it’s back to what we said. In these kinds of huge markets, you’re way out there on the risk continuum.

But when risk appetite is on—and risk appetite is on today—high-risk, high-return stories go at a premium, and this is a high-risk, high-return story with a big market. So in today’s market, in today’s environment, that could price extremely well.

Harry Stebbings

It’ll create an interesting case study with Groq. Who has the better outcome, risk-, and time-adjusted, right? I know this is such an annoying investor thing, but who has the better outcome, risk-, and time-adjusted?

Both took huge risks to start their companies. Both started ahead of the curve, right? Cerebras, I guess, even more, right? Would you rather take $20 billion in whatever—I don’t know—combination of cash and stock with weird taxes you got from NVIDIA, or ride the up-and-down emotional roller coaster of running a public company for a decade and trying to get liquidity and see if Morgan Stanley will give you a loan against your stock? We’re all on different journeys. I don’t know.

Harry Stebbings

I can tell you I’d rather sell to PE, like Salesloft did for $2 billion, and peace out.

Jason Lemkin

Well, that’s a different journey altogether, right?

Harry Stebbings

That sounds better.

Rory O'Driscoll

You guys, you’re doing the heights. I mean, yeah, look, equally, I think they’re pretty glad they didn’t sell Google to Yahoo for $1 billion, whatever it was. Look, when it works, you’re glad you—

Jason Lemkin

But what about Figma and Adobe? That didn’t happen. That one, I’d be pulling my hair out if I was Dylan. Dylan’s better than me, but I’d be like, “Farts.”

Rory O'Driscoll

Yeah. Yeah.

Jason Lemkin

Farts. Farts, farts, farts, farts. I don’t mean to be political, but, man, that was a rough stretch there.

Rory O'Driscoll

Yeah, but maybe they like what they’re doing. You know, lots of folks do, actually. Maybe they just—

Jason Lemkin

You do, but it’s tough when the team’s RSUs are not worth what they were, the options are struggling, and people are leaving to go to Anthropic. It’s just not a fun environment. No matter how great, it’s just not fun to run those companies, right?

Rory O'Driscoll

That’s my point. But you’re exactly right. When the momentum turns against you, you wish you’d sold, and when the momentum—I mean, Jensen’s glad he didn’t sell NVIDIA at any time along the way.

When it works, you’re glad you didn’t sell, and when it doesn’t work, you wish you had. It’s as simple as that.

Harry Stebbings

It’s a derivative of Elad Gil’s point. Which one are you, right? Be honest once a year at the board meeting. Be honest. Who are we? Are we NVIDIA, or which one are we?

And it’s easy on Twitter to think that you’re Jensen. We used to think we were Zuck, right? I’m the CEO, baby, right? Now we all pull on our leather jackets and think we’re Jensen, right?

Rory O'Driscoll

If you’re gonna be Jensen, I hate to tell you, but you’ve just got to be willing to eat a podcaster for breakfast once a week, because, oh my God.

Harry Stebbings

I was going to bring this up as a kind of—what did you guys think? I'm sure we all watched the Dwarkesh ones. What did you think?

Rory O'Driscoll

Where I thought he was excellent was in his comments on not having preferential status: “We’re here to sell chips. If you issue us a PO, we’ll sell you chips.” His relationship with TSMC—all that stuff, I think, was super grounded, and you just go, “Wow, that’s a world-class executive who has shipped gazillions of chips.” What you’re really talking about is the argument about China, correct? With Dwarkesh, he and Dwarkesh got into it, and it was a little bit—

Harry Stebbings

Well, yes, and when he was saying bluntly, “When you look at 2 of the largest frontier model providers, neither of them trained on your chips,” he provided a pretty cagey response at best, and then admitted, “We should have invested in them.”

Rory O'Driscoll

Well, I think there are a lot of things lumped into that. OpenAI definitely trains in part on NVIDIA chips. On Anthropic, I’m not as clear on what they train on. I’ve had mixed comments on that, but whatever.

I think his comment on not investing in Anthropic was that he couldn’t do it at the time because he didn’t have the capital, which makes sense. In 2022 and 2023, he wasn’t in the business of writing $30 billion venture checks. I thought he was very rational there. He said, “I made a mistake. I didn’t have the capital at the time. I wish I had.” Just bring them closer to you. So I didn’t think that was bad.

I think the problem with the China discussion is that there were 2 priors that neither party agreed on. When you have a discussion and they’re talking past each other, and you don’t agree on the ground truth, it’s just a waste of time. One is, how big an enemy do you think China is? Is it just a competitive trading partner, a competitor in the way that Microsoft and Apple compete? Or is it the new Russia, and we’ve got to not give them a single thing because we’re scared they’re going to nuke us?

The other thing is that Dwarkesh clearly thinks that frontier models are as dangerous as uranium, and Jensen clearly thinks that’s bullshit. If you don’t agree on those 2 things, if the question is, “Should we make it easy for China to build frontier models by selling them NVIDIA chips?”—which was the question—if you don’t agree on what you think about China and you don’t agree on what you think about frontier models, you simply don’t have a useful discussion, because neither of the nouns in the sentence have been defined.

Once you internalize that, you’re like, “Ooh, that’s 2 people talking past each other, and one of them simply doesn’t give.” That’s why that part wasn’t that useful, but it was kind of funny. It reminded you what semiconductor executives were like. When I started investing, a lot of the business was semiconductors: just hard-headed guys who say, “No, no,” and enjoy it. So I enjoyed it at that level.

It was a little bit of a culture clash between generations, and it was fun. But I thought he did—look, I can’t argue with the guy.

Harry Stebbings

No, I just saw that for the first time ever. It wasn’t an easy interview.

Rory O'Driscoll

No, it wasn’t easy. And give Dwarkesh credit. He tried to punch, and it’s really hard to punch someone who, A, talks his book, and B, has 30 years of knowledge.

Harry Stebbings

When you’re a 25-year-old podcaster.

Rory O'Driscoll

Absolutely. I think one of the things I like about it is that where you come from dictates how you approach things. We’re doing this in a spirit of inquiry because, to some extent, we’re all trying to figure out what we think, and talking things through often helps. So I often find I revise my priors based on the discussion.

But I’m not here as the CEO of a $5 trillion company. He’s here to talk his book. And let’s be clear: 30% of his entire market is in China. He wants to sell those guys NVIDIA chips, and there’s simply no argument on God’s green earth that’s going to convince him that he shouldn’t get that $40 billion in revenue from shipping chips to China.

He spent time with the president lobbying to be allowed to sell NVIDIA chips. If he’s had to do whatever it takes—and I shudder to think in terms of sucking up—to be allowed to sell chips to China, he’s not going to roll over and play dead because some 25-year-old said, “Maybe you shouldn’t.” He’d do the same to you, Harry—straight handoff, right in the face. It’s like, “Thank you, but no, big guy.”

Did I think he won the argument? No. But he knows how to hold his corner. So it was fun. I worked out and listened to it. I was like, “Whoa.”

Harry Stebbings

Boys, are there any other topics that we haven’t discussed that we should discuss?

10. Snap Struggles As AI Recentralizes

Rory O'Driscoll

We ran out of time to cover Snap, but I just don’t care.

Harry Stebbings

And for anyone listening, Snap cut 1,000 jobs, 16% of the workforce, and the stock popped 11%.

Rory O'Driscoll

They just need to figure out a convincing business model, and they haven’t. Amazing company at the start, and now it’s just drifting around and needs to figure it out.

Harry Stebbings

Master of stock-based compensation.

Rory O'Driscoll

Yeah. No, I mean, it’s going to be an advert for the dangers, to some extent, of dual-class votes. I used to agonize about this, and now I don’t.

Now my perspective would be, look, you had a chance to buy 2 social media companies with dual-class votes. If you bought Snap and if you bought Facebook, you’re 50% down on Snap and you’re 10× up on Facebook from the IPO. Shut up, take the check, move on. Turns out untrammeled power has good outcomes and bad outcomes.

Harry Stebbings

He did marry a supermodel, though, so it’s not all downside. There’s a positive note to end on, Rory.

Rory O'Driscoll

I hope everyone has happy marriages. There. That’s great. Good show.

Jason Lemkin

Rory, okay, I have one last one if we’re out of time, but I want to get Harry’s thoughts. The other thing is, Elon had retweeted something that went around many times: 91% of all AI unicorns are now in the Bay Area. Thoughts from London on this?

Rory O'Driscoll

Oh, nice one.

Jason Lemkin

How many are in Marylebone? But with 91% in the Bay Area, Project Europe, how are you thinking about that from London—the increasing concentration of AI unicorns in the Bay Area? This is from this week.

Harry Stebbings

I think some of the best AI minds, or the majority of the best AI minds, want to be in Silicon Valley. Quite rightly. You’re seeing the recentralizing of power back to Silicon Valley.

That said, I think you can still build unbelievably great AI companies, as we have done in London with Demis and DeepMind, and with Mati at ElevenLabs. I think it’s easier to be in Europe because, with your 91%, you also have 91% of the capital, and every other person on the street is a venture capitalist.

So I think you see this gluttony of cash combined with a gluttony of companies, which makes detection harder and makes winning harder. I agree with that. Sure, the majority of great AI companies are there, but I also think it’s harder.

For me, being one of the top 3 brands in Europe, I would rather be here with much less supply-side competition than there, fighting against Benchmark, Founders Fund, Andreessen, and everyone in between.

Rory O'Driscoll

Well, I have 2 comments on that. First, it’s the old Caesar quote: “I’d rather be first in a village than second in Rome.” That’s really what you’re saying. Though, of course, I would add that he was a killer psychopath and, really, not a good man. The evil that men do—we can go back to Shakespeare, but we won’t.

I think the real point is what you’re saying. It’s interesting: it’s 91% in the Bay Area. What this says is that companies are in the Bay Area to the point where the marginal advantage of being in the Bay Area is less than the marginal advantage of being in Europe and having access to a talent pool.

What it says is that the equilibrium point for indifference is now roughly at 90%, which is another way of saying, yes, to a rounding error, the Bay Area wins, but there are still some wins in Europe.

Harry Stebbings

And I think, actually, trying to be a startup, staying, recruiting, and maintaining that team in the Bay is next to impossible unless you have an egregious amount of money.

Jason Lemkin

No doubt. I wasn’t expressing an opinion; I was just curious, as a last point, what you thought of this from this week. I don’t have an opinion on this. Certainly, you can’t argue with the talent question, but there’s just a lot of complexity here as everything concentrates. Everything’s concentrating in everything.

Harry Stebbings

And also, Jason, just think about the competitive funding landscape. I do not mean this arrogantly, but there are just 1/10 as many competitive elements in Europe as there are elsewhere. I have so much respect for Silicon Valley early-stage investors. God, it’s fucking competitive.

Rory O'Driscoll

It is. Again, the equilibrium will be reestablished when the costs of being here are equivalent to the advantages.

Look, I think really what happened is that for a couple of years—and it gets back to Anthropic and everything like that—in that period when they first cracked the code at OpenAI of what could be done, the closer you were to knowing what was happening, the bigger the advantage you had. It was intrinsically a local thing.

You’ll look back and go, “Yeah, 10 years after 2022, knowledge will be widely dispersed.” But there was a period of 2 or 3 years where the knowledge was available tribally in hacker houses in San Francisco and wasn’t available widely across the rest of the world. That’s why you had this Cambrian explosion here.

It’s a point in time, just like the start of the internet. But, yeah, great companies in London.

Harry Stebbings

And just to be clear, I’m not a—was it a psychopathic serial killer? Was that what you said?

Rory O'Driscoll

Yeah, I know you’re not, Harry. I know you’re not.

Harry Stebbings

A slightly mediocre moderator, but not a killer. Right?

Rory O'Driscoll

Yeah. I mean, remember, Caesar killed a million Gauls. Let’s just keep score. Gratuitously.

Harry Stebbings

Not quite as good as ol’ Pop, but, you know, up there.

Rory O'Driscoll

Well, up there in the baddie category, yeah. Okay.

Harry Stebbings

Awesome, boys. Well done. But before we leave you today, let me tell you about Omni. It's an AI analytics platform, and it solves a problem every scaling company hits. Your team needs insights, not just data lookups, the stuff that really matters, and it's critical to get it right, like CAC payback periods and net dollar retention. For AI agents to act on your company data, they need your business context, your definitions, your logic, your permissions, and that's what Omni's governed context graph provides. Your data team defines it once, then anyone, your ops lead, your CFO, your PM, can ask a question in English and get an answer in seconds. Perplexity, Mercury, and DBT run on Omni. And 20VC listeners get a free three-week trial. Three week, very specific. Not a month, but three weeks. Go to omni.co/20vc. That's omni.co/20vc. After Omni helps you find the right customers, Checkout helps you close them. Over the past 15 years, Guillaume Pozas has led checkout.com through what he calls the velocity years, a period of hypergrowth with relentless product building. The lesson? High growth is a gift, but it demands ruthless focus. As his mother put it, "Play the game you're good at." For checkout.com, that game is digital payments, obsessing over data, chasing basis points, and compounding learnings over time. And that discipline is paying off. 2025, checkout.com processed over 300 billion in total volume, up 64% year over year, and returned to full year EBITDA profitability. They now support over 1,000 enterprise merchants globally, including 63 that process more than a billion annually with brands like eBay, Vinted, Amex, ASOS, and Temu. Guillaume's message though, it's pretty clear. They've earned the right to win anywhere. Now, they're investing in innovation across marketplaces, issuing financial experiences, and agentic commerce. If you want payments built for what's next, talk to the team at checkout.com. That's checkout.com. While Checkout powers the moment money changes hands, Invisible powers the people behind the work. Why don't we hear more real AI success stories from big companies? The models are insanely good, but implementation's the problem. It's really, really hard. There's data all over the place. There's legacy tech and manual workarounds. It's a Ferrari engine in a shopping cart. Meet Invisible. Invisible trains 80% of the top models and then adapts them to the messy reality of your business. Take the Charlotte Hornets NBA team. Invisible took years of game tape and analog scouting notes to go from uncertainty to a draft pick and Summer League championship win in weeks, not seasons. Get the data in order first, and suddenly AI can do almost anything for you in the enterprise. If you want AI that hits the P&L, go to invisibletech.ai/20vc.

Cursor Acquired for $60BN | Anthropic Hits $1TRN in Secondary Markets & Figma, Adobe, Canva Dead? | BidClub