[BidClub_]
20VC · · 92 min

20VC OGs: SpaceX Valued at $800BN & Harvey Raises $160M at an $8BN Price | Airwallex Raises $330M and The Battle with Keith Rabois | Netflix Acquires Warner Brothers | IPO Market Predictions for 2026: Anthropic, Stripe, Databricks and SpaceX

Harry Stebbings

Podcast ↗
TL;DR

  • SpaceX may be an extraordinary company and still be an unattractive purchase at an $800 billion valuation. Rory O'Driscoll described two exceptional businesses—rockets and Starlink—but roughly $15 billion of revenue growing about 30% leaves it above 40x run-rate revenue, with “a lot of non-obvious math” and an Elon premium embedded. His clean comparison: Anthropic could command roughly 20–25x while growing 300%, versus SpaceX at roughly 40x growing 20–30%; “it is possible to lose money on a great company.”

  • A single blockbuster listing could make 2026 a record IPO year without resolving the wider venture-liquidity deficit. Rory’s illustrative marks for SpaceX, Anthropic, and Databricks total $1.4 trillion; if VCs own slightly under half, roughly $700 billion could return against an estimated $2.8–$2.9 trillion of private venture value. Jason Lemkin predicts Anthropic and Databricks will list in the second half of 2026 if markets hold, while Rory warns that even enormously successful IPOs could price below their last private rounds.

  • Netflix’s proposed $82.7 billion Warner Bros. Discovery deal is the moment the internet-native distributor starts consuming the studios it already defeated. A roughly $470 billion Netflix can put nearly $100 billion on the table with less than 20% dilution, while sub-$20 billion Paramount needs debt and outside capital to make a similarly sized bid. Regulatory, political, and Hollywood opposition remain substantial, especially because Netflix could become the dominant buyer of creative work.

  • Tiger Global’s smaller, concentrated fund reflects a reset after 2021, while Naveen’s $500 million seed at a $5 billion valuation reflects the premium attached to repeat success. Tiger made nine deals this year versus more than 100 in 2021 and is committing 20%, roughly $400 million, of its own capital; Rory called money “a great truth serum.” Rory could write Naveen’s investment memo—“Once you’re lucky, twice you’re good”—but not yet demonstrate a clear $25 billion outcome. Earlier, cheaper rounds can also make a headline seed valuation different from the entry prices of early investors.

  • Harvey’s $8 billion valuation works only if exceptional growth remains durable and the company expands beyond legal AI. Reported discussion centered on roughly $150 million ARR, 300% growth, 98% logo retention, and 170% NRR. Jason’s case is to back an outlier that grew from $50 million to $150 million and could approach $450 million; Rory’s concern is whether the deceleration path is 10x to 6x to 3x or 10x to 3x to 2x. Harvey must also defend against model improvement and expand into labor or other professional services.

  • AI application moats remain provisional because model improvements can strengthen today’s leaders or invalidate their architectures. Harry emphasized implementation, workflow integration, and customer relationships; Jason’s counterexample was a step-function improvement in deep reasoning, from minutes to seconds. Rory’s middle ground was that domain workflows matter, but “product-market fit is a rolling feast,” requiring application companies to evaluate every new model immediately. LLMs can be easy to swap without necessarily becoming bad businesses: applications can route among several model APIs in real time, although Rory still expects high fixed costs to reduce the field to perhaps three or four viable providers. The Chinese open-source debate added a separate legal and security question: Andreessen’s 80% figure was described as inaccurate, while Jason and Rory agreed that cost, functionality, and government restrictions must be evaluated separately.

  • ChatGPT’s habit and memory moat is valuable but contested. Jason said users could move to Google within days if ChatGPT disappeared; Rory argued that memory, habit, product quality, and roughly 800 million users would create a meaningful void, using the test: “If you lost it today, would you go out and buy another one tomorrow morning?” OpenAI’s “code red” was framed as a need to concentrate on healthcare, codecs, and the consumer product.

  • Airwallex’s $8 billion round looks like a valuation dislocation, but geopolitical exposure has become part of the asset. At roughly $1 billion of revenue, Airwallex is priced at a quarter of comparable-revenue Ramp’s $32 billion mark; Rory and Harry attributed much of that gap to an Asia discount while distinguishing the Australian company from a Chinese-owned business. Rory would require a plan to eliminate its China operating exposure; Jason argued that 8x revenue is attractive and great founders should be trusted to resolve the risk.

  • Prediction markets may face an insider-trading and manipulation reckoning. Kalshi’s $11 billion mark and Polymarket’s $13 billion mark prompted discussion of anonymous traders apparently winning millions with potentially privileged information. Rory also warned about controllable sports micro-bets and predicted congressional scrutiny in three or four years.

Digest · the substance, structured for research

1. SpaceX’s $800 billion mark prices in more than excellence

  • Rory began with the quality before the valuation: SpaceX is “an amazing rocket company” and an amazing communications company through Starlink. But roughly $15 billion of revenue growing plus or minus 30% puts the secondary above 40x run-rate revenue, enough that he might sell rather than buy.

  • His concern is not that the company fails, but that cash flows must eventually justify an enormous Elon premium. “In the long term, everything…is a weighing machine,” and without that premium the stated price would look “pretty sick” in public markets.

  • The cross-asset comparison sharpened the point: SpaceX may receive roughly 40x revenue while growing 20–30%, whereas Anthropic could receive roughly 20–25x while growing 300%. SpaceX is profitable, competitively singular, and addresses a huge market, but Rory still saw “the great-leader premium” doing substantial work.

2. One mega-listing can manufacture an IPO boom

  • Rory’s scenario assigned $800 billion to SpaceX, $400 billion to Anthropic, and $200 billion to Databricks: $1.4 trillion collectively. If venture investors own slightly under half, roughly $700 billion could return—an extraordinary year, but only about 20% of his estimated $2.8–$2.9 trillion of private venture value.

  • Power-law math makes IPO counts almost irrelevant. Facebook made 2012 a “bumper year,” Alibaba created another towering bar, and one SpaceX listing at $400–$600 billion would dwarf numerous $2 billion offerings.

  • Jason, already owing Harry $50,000 after predicting an earlier reopening, doubled down: Anthropic and Databricks have incentives to list in the second half of 2026 if markets remain healthy. His previous error was assuming B2B leaders would resume IPOs before realizing that most had stopped growing, except the AI companies.

  • For Anthropic, Jason chose a $500 billion IPO mark based on projected revenue around $25–$26 billion at 20x. Rory chose roughly $350–$400 billion and Harry chose $420 billion. Rory’s condition was that the valuation works if extreme growth attenuates rather than collapses under sustained deceleration.

3. The open IPO window can shut as quickly as it appeared

  • Rory noted that equities are at elevated valuation levels not merely because indexes often reach nominal highs, but also on P/E and Shiller P/E measures. He described the rebound since the April 15 tariff selloff as the strongest bounce back from a bear market since 1982—evidence that sentiment can reverse just as quickly.

  • Even if the three headline companies listed at only $700–$800 billion collectively rather than $1.4 trillion, Rory said it could remain the best year ever. Yet late private buyers might still feel they overreached, demonstrating that a down IPO can return vast capital while exposing a bad entry price.

  • Anthropic appears unusually likely to use an open window because, despite its unconventional founding philosophy, Rory sees its financial conduct as “conventional, sensible”: lower burn, early convergence, and recognition that a capital-intensive company valued around $400 billion may eventually gain a strategic advantage from public-market access.

4. Netflix won distribution and can now consume the studios

  • Rory corrected the premise from Netflix “acquiring” Warner Bros. Discovery to Netflix hoping to acquire it amid Paramount’s hostile challenge. The decisive asymmetry is financial: roughly $470 billion Netflix can place nearly $100 billion on the table, while sub-$20 billion Paramount must assemble debt and private capital to do the same.

  • WBD’s board appears to prefer Netflix because certainty of close matters. Jason suspected management also dislikes the prospect of working for Paramount and may prefer a transaction that gives it more control over its future. A hostile tender could reveal whether shareholder and management interests diverge.

  • Jason added the multiple arbitrage: Netflix equity trades at a much higher multiple, potentially allowing it to revalue acquired revenue and make the transaction highly accretive. Rory’s more fundamental explanation was global distribution—Netflix can pay more for content because it can monetize that content across more customers. “The business model won.”

5. Regulatory risk turns on whether the market is streaming or entertainment

  • Netflix faces the sharper FTC argument because both it and HBO are streamers; Paramount faces more FCC exposure through broadcasting licenses. A narrow streaming definition could put Netflix above 30% and invite monopoly scrutiny, while Netflix will define the market broadly enough to include YouTube and broadcast television.

  • Politics further complicates the probability of close. Rory pointed to the Ellisons’ perceived White House ties, investors connected to Jared Kushner, and comments from President Trump. His deliberately gentle formulation was that the executive office’s “thumbprint” may enter the process.

  • Hollywood’s objection is less consumer monopoly than producer monopsony. If Netflix becomes the dominant buyer, actors and creators lose bargaining leverage and fear being pushed toward cheaper, standardized output—“grinding out rom-coms on a low budget somewhere in Romania”—even if viewers still have abundant entertainment choices.

6. Software-backed distribution keeps swallowing old industries

  • Rory situated Netflix within a 30-year migration from software that mainly counted corporate resources to software that owns industries. Google and Facebook captured roughly 70% of all advertising, Amazon swallowed a significant portion of retail, and Netflix now demonstrates the same mechanism in entertainment.

  • Banking may be next. Rory imagined Revolut recognizing that it had become Europe’s largest market-cap bank and buying a branch-banking business, analogous to Amazon buying Whole Foods. Autos may follow over a longer horizon.

  • His through-line was the compounding ability of software and the internet to “eat, entirely destroy, and consume” other businesses.

7. Tiger Global is buying back credibility with concentration

  • Tiger’s reported new fund is about $2.2 billion, similar to its prior fund raised in 2023. Jason contrasted only nine deals this year with more than 100 in 2021, reading the change as a shift from the 2021 IPO-a-day strategy toward concentrated exposure to a few extreme winners.

  • Rory gave Tiger credit for admitting through action that 2021 reflected hubris. The original strategy—late, concentrated investments in the best companies—had worked before the firm became overextended. The reset shows someone wealthy enough to leave instead choosing to “swallow my pride,” absorb criticism, and stay in the game.

  • The reported 20% GP commitment, roughly $400 million, mattered more than commentary. Jason said growth funds often commit around 1%, sometimes through loans or a small number of wealthier partners; Rory’s verdict was, “Money is a great truth serum. Don’t tell me what you think. Tell me what you do.”

  • Jason questioned whether 20% leaves enough leverage after LP obligations, but Rory distinguished late-stage investing: someone with $400 million of personal capital cannot credibly lead a $200 million round with a small check. At that scale the external capital buys access, while the personal stake reassures LPs that difficult decisions remain financially consequential.

8. Headline seed valuations can conceal earlier entry prices

  • Naveen’s $500 million seed at a $5 billion valuation made sense to Rory as a founder bet: “Once you’re lucky, twice you’re good.” Two successful companies create enough evidence for VCs to back the person, although Rory could write that memo more confidently than one proving a clear $25 billion outcome.

  • Harry’s caveat was that splashy seed announcements can follow earlier, cheaper financings. The headline $5 billion price may therefore differ substantially from the prices paid by investors in the first rounds.

  • Jason said rounds at 3% dilution or less can be attractive to early investors, but small ownership stakes may not be treated as meaningful valuation marks by LPs. The underlying question is whether a high headline price represents durable price discovery or simply a very small transaction.

9. Harvey’s metrics support the bet, but not every implied outcome

  • The discussion treated Harvey’s $8 billion financing as roughly $160 million for about 2% dilution. The operating figures cited were roughly $150 million ARR, 300% annual growth, 98% logo retention, and 170% NRR.

  • Rory’s outcome math was more demanding: a 3x from the $8 billion valuation requires roughly $24 billion of enterprise value. No legal software company had historically been worth more than $2 billion, while legal-data businesses such as Westlaw and Thomson Reuters were described as worth several to tens of billions. Harvey therefore needs substantial TAM expansion into labor and other professional services.

  • Harry argued that even category leadership may be insufficient; Harvey may need Uber-like dominance rather than an application oligopoly. The counterexample is Legora, reported around $40 million ARR and growing 10x, with investors viewing Harvey as the U.S. leader and Legora as increasingly strong in Europe.

10. Harvey’s valuation is a wager on the shape of deceleration

  • Jason’s simple case began with the chart: Harvey moved from roughly $10 million to $150 million in two years and from $50 million to $150 million in the last year. If retention remains exceptional and the company approaches $400–$450 million next year, “this is just the bet you do.”

  • Harry’s pushback was price, not company quality. If $150 million merely doubles to $300 million and then growth declines to 80%, an $8 billion investor may be paying roughly three years ahead for the public-market valuation the business has not yet earned.

  • Rory framed the unknown as a decay curve. Old SaaS growth often fell to about 85% of its prior rate; AI leaders instead begin at 10x. A path of 10x to 6x to 3x supports almost any price, while 10x to 3x to 2x still allows good prices but risks getting ahead of the business.

  • That leaves valuation risk as the central exposure: Harvey is already plainly a great company, so investors are not mainly underwriting whether it exists or finds customers. “It is possible to lose money on a great company,” because paying $8 billion for something eventually worth $4 billion is still a loss.

11. A model step-change could reprice every application moat

  • Jason warned that Harvey and peers could be “Jaspered” by GPT-6, GPT-7, or Anthropic 5. Replit existed for eight years before Claude 4 made its experience work, and Gamma for four; another capability jump could similarly create products whose architectures are native to newly instant, deep reasoning.

  • Harry defended enterprise stickiness through implementation, go-to-market execution, workflow integration, and relationships at firms such as Wilson Sonsini and Cooley. His point was that an application sitting above several models can benefit when the models improve rather than being displaced by them.

  • Jason’s rebuttal was that 3–5% improvement misses the scenario. Customer-service agents remain slow, difficult to configure, and imperfect; if five-minute reasoning becomes five seconds or one second, a new vendor could be 10x better despite incumbents’ integrations. “We haven’t even begun to see deep reasoning in B2B apps.”

  • Rory occupied the middle: domain workflows and customer commands create lock-in if applications keep pace, but a genuine reasoning or AGI step-change could invalidate current designs. “Product-market fit is a rolling feast,” so strong teams test new models immediately—“pizzas and late at night”—to discover what changed.

12. Swappable LLMs, Chinese models, and consolidation

  • Harry contrasted model “promiscuity” with cloud lock-in. Moving between major clouds can take two years and millions of dollars, whereas AI applications often call several APIs and route tasks among them in real time; Sierra’s architecture was cited as a constellation of models.

  • Rory rejected the leap from swappability to permanently bad economics. The hard-disk analogy actually demonstrates that brutally competitive commodity markets can shrink from roughly 30 suppliers to two or three; an airline-like model with huge fixed costs and easy switching can also earn money after sufficient consolidation.

  • The strategic response is to own the end user. If Anthropic, OpenAI, and Gemini are interchangeable at the API layer, Anthropic has reason to capture coding workflows and OpenAI to protect ChatGPT. Rory expects perhaps three or four scaled model providers if entry remains expensive—not ten continuously interchangeable businesses.

  • The Chinese open-source discussion added a separate legal and security question. Rory said Andreessen’s reported “80%” figure was not correct; Martin Casado’s clarification was that only about 20% of the companies’ usage was open source, and only about 20% of that open source was Chinese. Jason said inference costs may limit the savings and that security matters; Rory separated national-security and regulatory compliance from the ordinary business questions of cost and functionality.

13. ChatGPT’s habit moat is valuable but contested

  • Jason argued that if ChatGPT vanished, users would complain and then move to Google AI Mode or another model within days; he provocatively said losing Netflix might hurt more because its content is proprietary. A later exchange also suggested that Anthropic’s enterprise position could be stickier than a consumer interface.

  • Rory disagreed using a consumer test: “If you lost it today, would you go out and buy another one tomorrow morning?” He would immediately seek a replacement for ChatGPT and does not consider Gemini equivalent; memory, habit, product superiority, and roughly 800 million users create meaningful gravity.

  • Jason compared that stickiness to Yahoo Mail—real until Gmail became good enough. Rory’s reply was that displacement required years of Yahoo under-execution, while ChatGPT still offers a differentiated experience today.

  • Harry read OpenAI’s “code red” as acknowledgment that it had diversified too broadly and fumbled the prior year. He expects concentration on healthcare, codecs, and the consumer product. Rory added that rankings at the end of 2026 or 2027 could “lock down” trajectories for a decade after this formative period.

14. Airwallex’s discount mixes fintech economics with geopolitical friction

  • Airwallex raised $330 million at an $8 billion valuation, led by Lee Fixel’s Addition, with roughly $1 billion of revenue. Comparable-revenue Ramp is valued at $32 billion; Brex was cited around $13–$14 billion despite roughly $400 million less ARR, weaker growth, and lower profitability.

  • Rory attributed part of the gap to ordinary payments multiples and Ramp’s unusually strong execution premium, but agreed that a significant Asia discount remains. Harry’s formulation was starker: assuming similar revenue, growth, and margins, Airwallex trades at one-quarter of Ramp’s price.

  • Rory distinguished Airwallex from a Chinese-owned business. He described it as an Australian or international company subject to Australian law, while noting that employees or infrastructure operating in China are subject to Chinese law, just as operations in the U.S. or Europe create local regulatory exposure.

  • Harry reported zero churn from the controversy. Jason nevertheless treated Keith Rabois’s public data concerns as a new sales objection, possibly intended to impede financing or momentum: even if the charge is unfair, the company must make its answer “unimpeachable,” while accepting that some U.S. government contracts may be unavailable.

15. Airwallex exposed a real divide over what boards should control

  • Forced to choose, Rory preferred Airwallex at $8 billion over Ramp at $32 billion, assuming comparable growth. His proposed closing condition was that there be no paid employees in China within 24 hours, followed by relocating the service organization to another jurisdiction within the next 12 months.

  • Jason called that posture “beyond condescending”: Jack Zhang knows the engineering, legal, and geopolitical trade-offs, and great founders should be trusted to solve them. At roughly 8x revenue, he saw a market dislocation—“You could do worse than invest in a dislocation in the market.”

  • Rory refused the suggested mulligan. A board, he argued, must intervene around “strategic fatal-error risk,” especially anything that dramatically narrows future buyers or government contracts. Executives close to trusted employees may underestimate how quickly those facts become irrelevant after a geopolitical event changes the diligence standard.

16. Prediction markets are approaching their insider-trading reckoning

  • Kalshi’s reported $1 billion raise at an $11 billion valuation and Polymarket’s $13 billion mark prompted Jason’s deliberately caustic thesis: the exciting use case is wagering on confidential corporate information. Rory cited an anonymous prediction-market participant apparently winning millions by predicting Google’s top query across sequential days, inviting suspicion of privileged data access.

  • Rory warned of both insider trading involving arcane information and manipulation when a bettor can control the outcome, especially through sports micro-bets tied to a particular play rather than an entire game.

  • His forecast was a “cesspit of issues” followed by congressional hearings in three or four years, echoing the quiz-show scandals of the 1950s. Anonymous or crypto-based participation will face basic questions about customer identity, repeated improbable accuracy, and whether operators can distinguish informed forecasting from manipulation.

Rory O'Driscoll

Every time private market valuations came into contact with public market valuations, private market valuations were found wanting. And money is a great truth serum. Don't tell me what you think. Tell me what you do. It is possible to lose money on a great company.

Harry Stebbings

Airwallex at $8 billion or Ramp at $32 billion—which would you rather own?

Rory O'Driscoll

Jason, bail me out here.

Jason Lemkin

A VC condescending to tell the CEO how they're going to work it out—it's beyond condescending. The great CEOs will figure it out.

Rory O'Driscoll

I'm going to push back hard on that. Hard on that.

Jason Lemkin

Okay. But people are going to see you as condescending, Rory. I wouldn't push back.

Rory O'Driscoll

I disagree.

Jason Lemkin

They like you a lot now. Don't make them think of you as condescending. It doesn't help at the end of the year.

Rory O'Driscoll

No, I'll deal with that straight away.

Jason Lemkin

Don't push back. Take a mulligan and delete this section.

Rory O'Driscoll

No, I won't.

Harry Stebbings

Guys, it is so good to be back. Jason, I am loving this 20VC swag that we've got going on here. It looks fantastic. Rory, we've got to get you some, my friend.

Jason Lemkin

Hey.

Harry Stebbings

A reminder to founders and others out there: swag works.

Rory O'Driscoll

Swag works.

Jason Lemkin

It works.

Rory O'Driscoll

It's a good investment.

Jason Lemkin

Actually, it's a good investment.

Rory O'Driscoll

It is. Quality swag works. It's got to be good enough that you want to wear it. That's the test.

Harry Stebbings

Or play with it in the case—

Rory O'Driscoll

It works.

Harry Stebbings

—of a paddle bat. Hey, Rory.

Rory O'Driscoll

Yeah, it works.

Harry Stebbings

We see yours behind you. Oh, yeah.

Rory O'Driscoll

And the others are paddle batters.

1. SpaceX at $800 Billion

Harry Stebbings

Okay, we're starting with No. 1. SpaceX is pursuing an $800 billion valuation through a secondary sale. This is obviously not the first time we've seen SpaceX doing large secondaries. It's the first time it's been $800 billion. How do we think about this news?

Rory O'Driscoll

It's an amazing company. Let's start with that. It's actually 2 companies: an amazing rocket company and an amazing communications and Starlink company. Doing $15 billion, growing plus or minus 30% this year, that's a pretty hefty valuation. That's north of 40 times run rate for a company growing 30% this year.

Now, you could argue that maybe there's some extra Starlink growth buried in there, but I remember thinking, “Hmm, I'm not sure I'd be a buyer at that price.” And if I owned some, I might be looking to sell. I think there's a huge amount of Elon magic overlay, and so far that magic has worked, but it's definitely a lot of non-obvious math baked into the price.

Jason Lemkin

It may segue into the IPOs of 2026 and 2027. I think our jaws are going to drop with the IPOs.

One question—I guess it's a minor question—is that there are so many brands that will IPO. In 2025, we had no brands IPO. I mean, we had some good IPOs. We had some IPOs that traded weak, like Figma, but it's not like folks on the street knew who Figma or CoreWeave was, right?

Who knows what retail will do to SpaceX, Stripe, and street names like Databricks, which everyone talks about? A little facetious there. But I don't know what the retail premium is for a hot name, though it has boosted Tesla over the years for sure.

Harry Stebbings

But I'll push back, Jason. Does this secondary sale not actually just show the lack of need for these names to IPO? The fact that you're doing it at $800 billion further supports the discussion we had with Tom Tunguz about the lack of need for these companies to IPO. I actually think it's not going to be a good 2026 because of secondary sales like these at these prices.

Jason Lemkin

But if Anthropic really IPOs at the end of next year, it would just— You know, we haven't had an IPO like that almost ever. It will just change the course. And the amount of liquidity, the scale—I mean, who knows? If it's worth $800 billion or whatever it trades at, these are just not like the VC deals we used to do. Even just 2 of them.

You're right. Maybe Tucker's right. Maybe these guys never IPO. Eventually, they will want the capital, I think, and I don't think Databricks and Anthropic are hiding from an IPO. I don't think they're doing a Stripe. I think they've been very clear they're on a path to an IPO. We just don't know when Databricks and Anthropic will IPO.

Rory O'Driscoll

Some of these are destined to go public in a way that maybe Stripe chooses not to. I think the question will be: if you get locked into a high price on a secondary, even if it doesn't have an IPO block, even if it's entirely secondary shares, will you get into this weird dynamic where it doesn't feel like a win if the public markets don't think you're worth $800 billion and only think you're worth $400 billion?

I think that's one of the weird things about private rounds providing a high-water mark. Does it make IPOs, for lack of a better word, emotionally unattractive because it's not going to feel like the win you wanted it to be?

Jason Lemkin

Does anyone care anymore, Rory, if it's a down round?

Literally, does anyone care anymore? I think we've given up caring about a down IPO.

I don't know if they care, though.

Rory O'Driscoll

I think they do.

Jason Lemkin

What I mean is, seriously, going into 2026, of course they care, right? Especially without a ratchet. My sense is it's baked into the business model, right? For Harry, on the off chance Airwallex is worth less than he paid, okay, I don't think Harry's going to quit the business, right? I don't think he has a 3X ratchet on his investment.

I'm just saying, much like coming out of a hot accelerator, raising at $60 million post with trivial revenue, I think even seed investors—

Rory O'Driscoll

I think, Jason—

Jason Lemkin

—have internalized this, right?

Rory O'Driscoll

But you are right. It is in the business—

Jason Lemkin

Yeah.

Rory O'Driscoll

But you are right. It is in the business model in the sense that, if you think about late-stage investing—and obviously, I think we can stipulate that $800 billion pre is definitely late stage by any man's definition—you’re not running any other kind of meaningful risk except valuation risk. So you can cry like a baby when valuation risk bites you in the ass.

At one level, you're right. Not only is it baked into the business model, it arguably is the business model. Just like investing in stocks, the business model is that some go up and some go down.

So, yeah, I agree. It's not going to be, "Oh my God, we'll never do that again," but what it will lead to is perhaps the point in time at which there's either less capital or more circumspection in the private markets. If you continually price something at a high price, and you're continually wrong and it goes public at a lower price, at some point some adult in the room will say, "Maybe we should stop doing that and wait and buy them when they're cheaper," right?

And you're right, it's not the end of the world, but that's the— As I look at $800 billion pre for a $15 billion revenue business, as I say, in the absence of the Elon premium, I think that price would look pretty sick in the public markets. Now, because of the Elon premium, I freely admit it's just not knowable to me in the short term.

I think in the long term, everything, as we've said, is a weighing machine, and the cash flows will dominate. So they're going to have to do a lot of growing to get to that price.

2. The 2026 IPO Wave

Harry Stebbings

I want to expand this discussion to the broader IPO market, because we've seen not a huge amount of liquidity come back, and it hasn't been as exciting or as exuberant as we thought it would be in 2025. Two questions, as direct as can be: Will 2026 be the year of the IPOs? If so, what will be the catalyst to drive that excitement?

Rory O'Driscoll

We were just doing the math when you sent out the note and said, "What happens if SpaceX, Anthropic, and Databricks go public?" Just doing the math here, I got $800 billion for SpaceX, $400 billion for Anthropic, and $200 billion for Databricks. That's $1.4 trillion of market cap.

Let's say VCs own, on average, a little under half of that, plus or minus. That's $700 billion of money returned to VC. That would obviously, to state the banal, be a pretty darn good year. It's sobering to realize it's only about 20% of the total private FMV, because the total private FMV of venture is something like $2.8 or $2.9 trillion.

It's a big chunk of the deficit. It's not all the deficit, but it would be a great year. And that's plausible. The weird thing about power-law math is that it only takes 1 or 2 of the top of the power law to go public to dwarf 10 little $2 billion pre-IPOs. It's all lost in the noise of the banker fee if SpaceX goes public at anywhere from $400 billion to $600 billion.

One of these years will be a bumper year for IPOs, and the thing that will make it a bumper year is, just like 2012. You don't remember, but 2012 was a, quote, "bumper year for IPOs." Why? Because Facebook went public. The year—I can't remember if it was 2018—was, quote, "a bumper year for IPOs" because Alibaba went public, and both of them were huge and ginormous. If you look at the little bar graphs, it's like, "Oh yeah, that was the year of Alibaba."

Harry Stebbings

What is the Facebook or Alibaba of 2026? Is it SpaceX, Anthropic, or Databricks?

Rory O'Driscoll

Oddly enough, any of them. Actually, that's not true. At $200 billion— I can't remember Alibaba, but SpaceX and Anthropic would be bigger than both. Facebook was around $70 billion or $80 billion. I remember Alibaba being bigger, a couple of hundred billion, but yes, any of these would be bigger.

At some point in 1 year, 2 or 3 of these will go out at the same time, and it'll be like the mother of all years.

Harry Stebbings

Do we not feel pretty good actually looking at this? When you add an OpenAI on top of that into H1 2027, which is kind of where people are projecting it to be, you're looking at 18 months of a pretty phenomenal funnel and liquidity.

Rory O'Driscoll

If the markets stay strong and that happens, the overall return to equity will be extreme—you know, the return to venture will be extremely good. Just to say it out there, because I am Mr. Debbie Downer: Markets are at an all-time high, both in index terms—which is meaningless, because they're often at an all-time high—but also in terms of valuation on a P/E, any kind of Shiller P/E basis.

In a much more normalized market—going back to perhaps disagreeing a little about something, Jason—if all these companies went public and every one of them was significantly down on the last round, let's just say instead of being worth $1.4 trillion, they're just worth $700 billion or $800 billion, it would still be the best year ever. Still, 3 amazing companies going out.

You could still have a weird feeling of, "Maybe I shouldn't have bought at $800 billion pre, or even $400 billion pre." That's the dilemma here. I think, net-net, it would be good overall. The capital would come back, all the prior rounds would make money. You might just have this odd phenomenon of people going, "That last price was a little reachier than the public," but you're still in good shape.

Harry Stebbings

Jason, do you want to lose some more money? Do you remember off that—

Jason Lemkin

I already thought this was going to happen at the end of 2024. I already owe you 50 grand plus interest.

Rory O'Driscoll

Wow.

Jason Lemkin

I was too optimistic. I really thought that we would have a blip in 2023. And in 2024, all the B2B leaders would come roaring back because growth stayed high for a while through 2023. So I thought 2024 was going to be—I thought we were going to have an IPO a day again because of the unicorn backlog. Little did I know they would all stop growing, except the AI ones.

Hopefully, the interest rate's like 1%, like Elon's loans. We'll have to figure it out, but I'm good for it.

Rory O'Driscoll

If your plan involves assuming we ever get to 1% interest rates again, I'd abandon that plan because—

Jason Lemkin

I can double down on the bet.

Rory O'Driscoll

Yeah.

Jason Lemkin

The bet now is that it has to happen by the end of 2026. It is tight, right? I think I'll take that bet. I'll double down on it. I think Anthropic and Databricks have an incentive to IPO.

Rory O'Driscoll

Yeah.

Jason Lemkin

As best I understand it, they have an incentive. As Rory said, times are great. Unless the market crashes for some reason, I think they're going to do it when the timing's perfect. I think they're both going to IPO in the back half of 2026.

Rory O'Driscoll

I do agree with you, because I think, oddly, Anthropic has been obviously very different in terms of its founding story, effective altruism, and angst about AI, but remarkably sober-minded and sensible and kind of mainstream in terms of the financial plan.

All along, they've stated they're going to be more efficient and burn less. They've talked about converging early. It feels like they will very sensibly go public when they can, because in this kind of capital-intensive business, I think at some point there will be a strategic advantage to going public.

So, logically, it would be the right thing to do, and they've been pleasantly more logical than the other player in this space, and more small-c conservative, which is ironic, as I say, given the philosophical approach there. But in terms of finance, they've been conservative, and I think they will be. Conventional, sensible, and mainstream maybe is a better way to put it than conservative.

The conventional thing to do when you're valued at $400 billion and you need to raise a lot of money would be to go out and access the public markets at scale.

Harry Stebbings

Jason, I'm keeping my gain. No deal. I think you're right. I think 2026 is when they go out.

Jason Lemkin

Wow.

Harry Stebbings

I do—

Rory O'Driscoll

Harry's pocketing his 50 grand and moving on.

Harry Stebbings

No.

Rory O'Driscoll

I love it.

Harry Stebbings

I'm pocketing my 50, baby.

Rory O'Driscoll

But also remember, I saw it in one of the chart guys I follow: from April 15, when we had the tariffs and the implosion, this has been the strongest bounce back from a bear market since 1982. It's been the biggest jump, right?

So where everyone's head was in April versus where everyone's head is today, you should remember that it can go the other way just as quickly. Everyone can think they're filing, and then something can go wrong.

You're right: when equities are at an all-time high, the window is open and you know you're going to need more money, maybe you should think about grabbing the moment.

Jason Lemkin

But if Anthropic is predicting its revenue is going to triple next year, it's difficult to imagine, even in an amazing market like today, that you're going to get a much better multiple than in early 2026, right?

Rory O'Driscoll

No.

Jason Lemkin

Will it triple from $26 billion to almost $80 billion? Maybe. But just the fact that you've got a 3X at that scale kind of in the bag, that's a good time to IPO.

Harry Stebbings

Jason, baby, I got a new bet for you before we do the BFD. The new bet is Anthropic going out. What price does it go out at? And Rory, you're involved in this bet.

Jason Lemkin

Some chance.

Rory O'Driscoll

I think there's talk about a private round of $300–350 billion. Again, looking at public information, when you look at the growth rate when they were doing the $170 million round, you can see the math working. So you could easily justify a $350 billion-type valuation in the public markets. That doesn't seem crazy.

I mean, it's amazing as a result, but Jason's right: 1 more year of 3X growth. The thing that would derail all this stuff, as we've all said, is sustained deceleration. But if the acceleration we're seeing attenuates but doesn't collapse, then that kind of number is totally plausible.

Jason Lemkin

I'm going to say $500 billion, which I think ties to the round.

Harry Stebbings

I'm writing this down. Jason, $500 billion.

Jason Lemkin

Yeah, and I'll tell you why. I think you've got to be careful to use ARR math when it's really forward growth, most likely, right? But if they do hit $25–26 billion next year, if it's in the bag, doing simple math, 20 times is what Databricks does. It's not even that high for a public multiple, 20X, right? For the top 10%.

So 20X times $26 billion is $520 billion. I'm going to guess a $500 billion IPO, right? It ties to a mediocre deal in this round, right? But still a good IRR, right? If they raise it at $300–350 billion, the IPO at $500 billion, that's still a pretty good IRR.

Rory O'Driscoll

I mean, if you think about the 2 conversations we've had, and I'm disappointed because I had all my homework done on SpaceX and we blew past it. But if you think about it, we're basically saying 1 company is going to get roughly a 20X–25X multiple, going 3X next year, down from 10X this year—10X and 3X.

And the other one's going to get 40X revenues, growing at—it grew in the mid-20s this year, but probably a little more next year because of Starlink, at 40%. So it's just interesting. We talk about these things in isolation, but when you zoom out and look across, would you prefer to buy the thing at 20X growing 300%, or the thing growing at 20–30% at 40X?

Maybe it does speak to some kind of premium. To be fair to SpaceX, it's an N of 1, it is profitable, it doesn't have a direct competitor, and it has a huge market. But I think a lot of that is just the great-leader premium.

Harry Stebbings

Rory, give me a price.

Rory O'Driscoll

Like $350 billion. I don't know. I'd put it at $350–400 billion. I'm not spending a ton of time on it.

Harry Stebbings

Okay, we're going to put $350 billion down for Rory. I'm going to go for $420 billion.

3. Netflix Eats Hollywood

Okay, the big fucking deal: Netflix acquiring Warner Bros., $82.7 billion. First question is, how do we analyze this? Second question is, will it even go through? How do we think about this?

Rory O'Driscoll

First of all, you're right. You say Netflix acquiring; Netflix is hoping to acquire, but with a lot of opposition from Paramount in a hostile bid. I think the big zoom-out venture comment is: Netflix won. They ate the media industry. Their market cap is $470 billion, and the biggest studio is sub-$200 billion. Comcast is, what, $100 billion. Netflix won. They can ingest this bite—it's less than 20% dilution—and keep powering through. That's the zoom-out comment.

They come to the table here, and it's just so interesting how outmatched the other player is, with 1 caveat that will come. Netflix is, call it, a $470 billion market cap company putting just under $100 billion on the table. Paramount is a sub-$20 billion company putting $100 billion on the table, relying on debt plus PE investors like the Kushner fund and a bunch of other funds to bridge the difference.

Those are very outmatched competitors in this hostile bid. Again, going back to the big picture, the outsider—the venture-backed company—has basically a far better business model than any of the studios and is eventually going to, as you're seeing here, start to eat them. That's the big picture.

Harry Stebbings

Do you think it will go through?

Rory O'Driscoll

I mean, there are so many dimensions to “going through.” First of all, it's clear the board of Warner Bros. Discovery—which is their ticker name, WBD—wants to do this deal. Paramount had put the thing into play, but the board's dynamic is: what do you do as a public board? You say, “What's the certainty of close? Hmm, the person who has $500 billion, plus or minus, is more likely to close than the person who has $20 billion, plus or minus. Hmm, I'll go with him.”

So they want to get that deal done, but you have a whole bunch of regulators, and then you have political overtones. We should also put in Hollywood artists, because they're fun, too.

From a regulatory perspective, Netflix is much more of a concern for the FTC because of monopoly power—alleged monopoly power—because HBO is a streamer and Netflix is a streamer. Paramount is much more of a concern to the FCC from a broadcasting-license perspective. Both of them have to go through everyone, but there's a different regulatory pushback for each of the 2.

All other things being equal, you'd say it would go through. It's the classic thing. If you do the narrow-in analysis—oh my God, if you just focus on streaming, then Netflix is big. HBO puts them a little bigger. They're over 30%. Everyone wrings their hands and says it's over.

Netflix will say, “Zoom out. Call this entertainment. Take into account YouTube. Take into account broadcast TV. We're still teeny-tiny, pushing through.” That's a known regulatory issue, and it'll get debated and litigated.

The 2 other things we haven't talked about, both of them are fun in different ways. One is that more than ever, you have political overtones here. Obviously, the investor base—even leaving aside Kushner—you obviously have the Ellisons, who appear to be pretty tight with the White House. You have Trump already commenting a little bit on the Netflix side. So you do have the thumbprint of the executive office probably coming in on this one, he said gently.

And the really fun one is Hollywood hates the Netflix deal. You have to ask yourself why. It wasn't obvious to me until I did some reading, and I thought, “Yeah, Hollywood is all about the creators.” It's not monopoly power screwing the consumers; it's monopsony power.

If Netflix becomes the biggest single buyer of content, then if you're making content—which is what Hollywood does—the media buyer for Netflix becomes the most important person in your life and your biggest customer, and they correctly hate that. They'll say it's because it'll be bland and boring content, and it will be, because corporations are more boring with content than individuals.

But let's get real: Netflix will exert its power not to overpay. It's always interesting because the regulatory approach kind of really focuses on the damage to the consumer, and that's not the case here. But if you're a producer of content—name me a famous movie star—you hate this. You're like, “I do not want Netflix to have all that leverage on me, or they'll have me grinding out rom-coms on a low budget somewhere in Romania, and I'm not going to like that. It's not going to be fun.”

Harry Stebbings

I'm sure Leonardo DiCaprio is terrified about that. Yeah.

Rory O'Driscoll

Absolutely. He might have to go down into single-digit millions for those 20-minute movies. There are so many ways to pile on that one. But, yeah, interesting.

Jason Lemkin

A small thing that's interesting to me is the CEO, Zaslav. Is that how you pronounce his name? David Zaslav?

Rory O'Driscoll

Yeah.

Jason Lemkin

Clearly, he does not want to do the Paramount deal. It is interesting, and if you've gone through M&A, you really want to have 1 real and 1 fake offer—at least 1 fake offer. You can tell corporate development you have another offer.

But if you've ever had 2 as a founder—and I've done both, had 2 and had a pseudo-offer—when you really have 2, you really think about what you want to do with your life more. The interesting dynamic is I just really don't think he wants to go work for Larry Jr. at Paramount. It's confusing whether he'll make more money under that deal or less, versus a deal where he gets to control more of his destiny and some of the assets stay on as a separate company.

As a founder—and he's not a founder—as a founder, I would take the Netflix deal in a heartbeat. And I do think that is some of this: “Hey, go away, Paramount,” because management doesn't want to do the deal. So it'll be interesting when they do a tender—the hostile offer, right?—if 51% of the shareholders take it, because there may be a divergence of interest at the margin between the 2. But I don't think management wants to do this Paramount deal.

Harry Stebbings

I thought it was interesting David Allison said the board hasn't even responded to Paramount's offer. No response at all.

Rory O'Driscoll

It is clear they were talking earlier. It's clear, as Paramount said, they put the deal into play, and then the company that you're trying to acquire kind of mentally gets its head around selling itself, and then it can just decide to sell itself to someone else.

Again, going back to the big picture here, someone else has a $480 billion market cap and you don't. That's what's happening here.

Jason Lemkin

At the margin, I don't know how important it is, but Netflix also has a revenue arbitrage. It trades at a much higher multiple.

Bill Gurley

Yeah.

Keith Rabois

That can come under a little pressure when you buy a large asset, but it still remains true. Netflix's equity is cheap, relatively cheap, to buy this asset, and they may be able to inflate all the assets to a 10X revenue multiple, right?

Rory O'Driscoll

Yeah.

Jason Lemkin

If they do that, this is an incredibly accretive deal for them versus the most expensive deal Paramount could do, even if it's do or die, right? The revenue arbitrage is not to be understated in a deal like this.

Rory O'Driscoll

Though it is interesting, the Netflix stock ticked down slightly, but I think that's a little short-termist. I think, even aside from the revenue arbitrage, the beauty of Netflix is, again, zooming out, because you have global distribution, you can monetize content better than anyone else. You can pay more for content because you've got more people to sell the content to. It's just that simple. The business model won.

But I actually think the fun question you asked, Harry, was what other examples of this will there be? You know, this is the digital world taking over old-school industries. Yet another old-school industry has rolled over and died, and I think that's a super question. You know, I was thinking about it a lot because—

Harry Stebbings

What are the other industries?

Rory O'Driscoll

I think there's a bunch, and that's the interesting thing. When I started in the '90s, IT was exactly that. It was information technology, and most of it was sold to corporations to effectively account for their shit. You had computers to count people, to count money, to count—That's all it did, hence the name. I mean, IT, information systems, those boring names. And that's still the bulk of core IT investing. You have software to run B2B.

But the amazing thing, this is to your point, Harry, is that only over the last 25 years, digitization has eaten a couple of industries. The obvious big one it ate was advertising. Facebook and Google have 70% of not just digital advertising, but all advertising. Thanks for playing, newspapers. Thanks for playing, broadcast TV. They won.

So from '95 on, venture-backed Silicon Valley startups ate the advertising industry and took all the money. And now you're seeing the second big one. In Mad Men, it looked like it was fun to be in advertising in New York. And now little tech weenies in Silicon Valley—we just did it to Hollywood. Thanks for playing, guys. I know you have these cute studio businesses, but we've got distribution on the internet. We've got our algorithms for predicting which content works. We're going to take over your sexy Hollywood business too.

And so you're right, Harry. What's next? I was thinking it was a great question. To me, you actually talked about it last time: the next one is fintech. At what point does someone like Revolut say, “Oh, I woke up and realized I'm the largest market-cap bank in Europe. Maybe I'll buy one of you branch banking thingies,” like Amazon bought Whole Foods? Because that's the third one that got swallowed. Retail got swallowed.

If you think about it, they swallowed advertising completely. They swallowed a lot of entertainment, not all of it. They swallowed a good slug of retail. Probably banking's going to go that way, and at some point maybe auto. It's just amazing, over an elongated period of time, the power of digital, software-enabled technology in conjunction with the internet to just eat, entirely destroy, and consume other businesses.

4. Tiger Global Resets

Harry Stebbings

Before we move to some of the biggest AI deals that we've seen go down, there are a couple that I thought were super interesting, and you can choose. Tiger's new $2.2 billion fund is downsizing. I know it sounds crazy—$2.2 billion is a huge amount of money—but from the lofty heights of '21, it's a big change. Anything of note that we should think about here?

Jason Lemkin

You know, I thought the most interesting thing in the reporting was that they did raise a prior fund of about $2.2 billion in '23, right? They only did 9 deals this year versus 100-and-some-odd in 2021.

I think what's just interesting is that this is a super-smart group of folks. I think both strategies are of the moment. I actually think the Tiger strategy in 2021, when we had an IPO a day and every recurring-revenue company could IPO at a couple hundred million in revenue, growing 50% with 140% NRR, was the strategy of the day.

Now Tiger's seen itself re-inflate with OpenAI and others, right? It's seen everything we've talked about: the huge growth, the unlimited growth for the winners. And now they've tilted to that strategy. I think growth investors probably should tilt to the strategy of the day.

But I just think it's interesting, and we'll see what the 2029 fund does. Doing 9 deals at that scale is pretty tight, right? It's a pretty concentrated focus, isn't it?

Rory O'Driscoll

It is. I give them credit. It takes a lot. Look, the guy running that has got billions of dollars; he doesn't need the grief. It speaks to a love of the game that he says, “I'm going to swallow my pride,” because I don't think he's going to say the '21 fund was a success or the right idea.

You might get bailed out by some of your big investments like OpenAI, but you don't look back. We all have periods as investors that we don't look back on with pride, where you go, “Mm, I got that wrong.” And I'm sure a team as smart as Tiger Global looks back on 2021 and says, “That was hubris, and I was wrong.”

I give him huge credit for not going, “And now I'm just going to take my $2 billion, $5 billion, and stay in my house in Palm Beach or whatever. I'm going to get back in the game. Raise a smaller fund. Take the little bit of knocks from the knockers that say you're wrong, and just do my job and do it well.”

Rory O'Driscoll

So I think it's great. All credit to them. What you're seeing, which is always true, is that the original strategy they had—doing the very best deals, doing them late, doing them with concentration—was the right strategy. It worked with JD.com in 2012 or whatever it was, when they did that in China. It worked early on in internet land. They got carried away, they made some mistakes, and now they're sobering up, flying right, and staying in the game. All credit to them.

Jason Lemkin

And a 20% GP commit is not small. Twenty percent is at the limit where it's almost not worth doing. You might as well just direct invest. You're just not getting enough leverage on your money, I've always thought. Twenty percent was a weird number, right? Assuming you have 20% carry, right? You're not making enough money. All the grief of having LPs isn't worth it at 20% GP commit, is it?

Rory O'Driscoll

Well, it is, because you need the bigger check size. At $400 million, you wouldn't be a late-stage fund.

Jason Lemkin

Yeah, you get to write bigger checks, but you don't make so much more money versus investing. In theory, you don't make enough money for the grief, I think.

Rory O'Driscoll

I understand what you're saying. The math is easy. You can do it in your head. It roughly doubles it. You have 20% of your money, and then you have 20% carry. So if you 2X the fund, half your gains come from your capital and half your gains come from your carry. It's easy math, ignoring fees, just for a second.

Jason Lemkin

Yes, but in theory—no, you're right. This breaks down on late-stage funds. For early-stage investing, I actually think it's worse because it's harder to get into the deal, right? It may be the opposite for growth, right?

Rory O'Driscoll

Jason, you're exactly right.

Jason Lemkin

Yeah.

Rory O'Driscoll

If you had $400 million of your own and you were doing an early-stage fund, becoming an $800 million fund to take other people's money, this is silly. You're exactly right.

I would argue, though, on the late-stage business, there's a critical mass. You can't show up at a $200 million round with a $30 million check and say, “Pick me.” That's just not a thing. He's sizing for the business at hand, and I agree. I think every LP probably draws a huge amount of comfort from that check.

I personally think, especially when you're dealing with wealthy people, where they put their money is what you want to know, and I actually think LPs don't focus on that enough. I find it enormously reassuring that someone cared enough to write 20%. And when things get tough, or when you're trying to, God forbid, get your capital back versus play to win, that $400 million will keep these guys focused.

So I think it's great. Again, everything about that is good. I hadn't picked up on the 20%, but that's awesome.

Jason Lemkin

For folks who might be founders or others, that's a lot for a fund, right? I think the average growth fund is about 1%, and even that can be manipulated in terms of where you come up with 1%. Maybe one or two of the partners that are richer are putting the money in. It can be from loans. It can be from cashless things.

Twenty percent is, to Rory's point, you've got to love the game if you're that rich. Even if you had a rough couple of funds, you've got to believe this is worth it, right?

Rory O'Driscoll

Agreed. You can't believe you're good.

Jason Lemkin

This is so much of the fund.

Rory O'Driscoll

And money is a great truth serum. Don't tell me what you think; tell me what you do. Do you love the game? That's one question. Do you love it at $400 million? That's a very different damn question. And clearly they do. At $400 million, you're in because you're in.

Harry Stebbings

I also think it pains him greatly that their brand has been tarnished in the LP community, and he will do everything possible to make other people money again. I think he is wounded and an incredible dude, actually. I'm a big fan of Chase.

Rory O'Driscoll

Totally.

Rory O'Driscoll

I've never met the guy, but that's how I read it. Absolutely.

Harry Stebbings

Yeah.

Rory O'Driscoll

There's a point at which you're not doing this for the marginal money; you're doing this because you like it and you like to do it well. You take money from LPs. You value that relationship. I totally agree. I think they'll be a highly disciplined, highly focused machine.

5. The $5 Billion Seed

Harry Stebbings

Next, Naveen, Databricks' head of AI, announces his $500 million seed round at a $5 billion price. We actually discussed this months ago, well before it was announced this week. I don't know if you remember it. It was just announced this week and caused a stir: the $500 million seed. Anything to discuss that we haven't?

Rory O'Driscoll

Sure. I can't remember who wrote the book about another entrepreneur where it perhaps didn't apply as much, but the title was Once You're Lucky, Twice You're Good. And Naveen's done it twice. You just look at that and go, at this point you can draw a line. And once you can draw a line, the VCs are going to just show up en masse. It totally makes sense.

And everyone's doing their “I'm delighted to back Naveen” tweet yesterday morning. It is true to some extent. There's such a lot of data in someone who's been successful twice. You just gotta look at it. I can totally see how you can extrapolate to that.

What kind of seed deal at $5 billion can give you a 3–5x return? I don't have clarity on that. It's a lot of enterprise value to be created. I know how I'd write that investment memo if I was writing it for my colleagues. It would be, “Once you're lucky, twice you're good. This guy's done it twice. Back him.” That's different than saying, “I see a $25 billion outcome here.” I just don't know it.

Harry Stebbings

The one lesson that I've learned from a lot of these deals, whether it's Thinking Machines or this or some others, is that although it's $500 million at $5 billion on the sticker price when it's announced, there are actually several rounds before that that Sequoia and Andreessen have done that are considerably cheaper. And the $5 billion sticker seed is actually very different from the first 2 rounds before that that no one's picked up on.

Jason Lemkin

You know, it's funny. When you can get a round like this for 3% or less dilution, I'm super in favor as an early investor, right? It's a great deal. Especially if we no longer care about the downside risk. If we no longer care if it's worthless, I think it's an amazing deal.

The funny thing is, when I started investing, I was taught—in fact, I was told by one of my anchors—these kinds of deals you can't recognize the markup. It's too small. I had a deal like this during the 2021 boom. It quickly got marked up to $3 billion by a good investor, but it was 2% of the company, right, to get into the company, and I never recognized the round or marked it up. My LP said, “You can't do it.” I guess here it's $160 million. That's not chump change, is it?

Rory O'Driscoll

No.

Jason Lemkin

But it's so small. It's so small. Is it a real valuation? I don't know.

Rory O'Driscoll

Well, just to push on that, right? $160 million over $8 billion is 2%, right? I mean, let's just do the math here. At an $800 billion valuation for SpaceX, 2% is $16 billion. They're not doing a $16 billion secondary. We just took that valuation seriously half an hour ago.

And the interesting thing is, again, we've said it a million times. At a 3x, it's $24 billion. There's no legal software company that's been worth more than $2 billion ever, so that's 10x that. The data companies, the Westlaws and the Thom Vests, are worth more than $2 billion. They're, you know, what? Fives and tens of billions, so still less than this. This implies huge TAM expansion into labor and, if you go on the website, huge TAM expansion, I'm sure, beyond just legal into other professional services.

Harry Stebbings

That's when you have to believe, if you're going to see this being attractive from here.

Rory O'Driscoll

I think, yeah, you do.

Harry Stebbings

And you also have to see market dominance, I think.

Jason Lemkin

Yeah, you have to be the winner. This is not the prize for number 2 in the world.

Harry Stebbings

But not only the winner, though, Rory. I think, actually, market dominance in a way that Uber and Lyft is. You can't have a cloud competition there where 4 players take— But there are so many, and there are so many fragmentations of this. You know, you did GC AI, I mean, so— I think you need to win the whole market.

Rory O'Driscoll

Yeah, I agree.

Harry Stebbings

But also not the unbundling, Rory. As I said, we're seeing this vertical unbundle into lots of different niches that we're both in and invested in. I think you have to believe they subsume all of those and be able to expand broader.

Rory O'Driscoll

And the theory would be: first, marquee customers in law firms in particular. That's the investment thesis. Top-of-class metrics.

6. Harvey's $8 Billion Bet

Harry Stebbings

The challenge to that, though, is LawGora are $40 million in ARR, so a lot less, but they're doing 10x year-on-year growth on a smaller base, admittedly. But they are doing very well in Europe. LawGora's European progress can't be underestimated, and really the analysis from investors today is that Harvey have won the US and Legora are winning Europe.

Jason Lemkin

I think we're making it too dramatic for VCs. Here would be my analysis, okay? If Harvey went from $50 million to $150 million this year, and $10 million to $150 million in 2 years, I would just do some classic math and draw the chart out. In my entire life in B2B as a founder or investor, every time we try to be too negative on TAM, it bites us.

Now, we talked about— What did we talk about in Live in London? What did we call it?

Harry Stebbings

The TAM Trap.

Jason Lemkin

The TAM Trap. I believe in it. I'm living it with my portfolio.

Rory O'Driscoll

But we agreed that those were older companies. I agree.

Jason Lemkin

When you go from $50 million to $150 million in 1 year, and if the last quarter—the last 4 months—show that the growth is consistent so you can see a path to $400 million next year, this is the same multiple we just talked about.

Rory O'Driscoll

But Jason, you could extrapolate this out for infinity.

Jason Lemkin

But that's our job as investors: to find outliers. Here's my point. Listen, I'm a simple man, okay? If this growth is durable, if we're seeing signs that it's durable, if it really has 170% NRR and 98% logo retention and it's accelerating at 150, this is just the bet you do.

Jason Lemkin

You don't pull your hair out or have lengthy dinners about supplanting labor with AI and whether it'll be 10 trillion tokens. You just do the deal if the revenue is durable and it triples from 50 to 150 in 1 year. The guys who got in at $3 billion at the beginning of the year got an insane deal. It's just not that complicated. Rory and I can go back to old SaaS spreadsheets and justify this deal.

Harry Stebbings

But you cannot assume that it will grow ad infinitum at the same growth rates.

Jason Lemkin

Great, then step out of venture and return your last fund—

Rory O'Driscoll

But no, to be fair—

Jason Lemkin

To your LPs.

Rory O'Driscoll

No, that's not—

Jason Lemkin

This is pretty simple math.

Rory O'Driscoll

No, but your big-picture point is a fair one. Let's just talk about the numbers behind it, because, Harry, he's not saying the same thing. What we're basically saying is it grew 10X last year and 3X this year. This is amazing growth. Lean in. That's the thought process that Jason's saying.

The only argument against it is that 3X this year is slower than 10X last year, so probably next year is 2X. It all boils down to the same thing: multiple companies doing 10X growth rates is not something you saw in SaaS. One of 2 things is true: either they'll have that kind of 10X growth rate and then decline with the same rate of decline as SaaS companies, in which case they'll continue to grow for a long period of time, or they'll grow fast and they'll slow down fast.

If it goes 10X, 3X, then 50% growth, you might have been wrong. That's the known fear. I agree. You lean into the things that are working, because I think the alternative, which is trying to be clever and do things that aren't working that might work in the future, is just too hard.

Harry Stebbings

So help me understand this. I totally get 10X to 3X. Then let's say it does 2X. So 150 turns into 300. Fantastic.

Rory O'Driscoll

Agreed.

Harry Stebbings

Then let's say it goes to 80%. That 300 is now—

Rory O'Driscoll

Yeah, Harry, that's exactly the right math to do.

Harry Stebbings

Okay, great.

Rory O'Driscoll

So—

Harry Stebbings

So then, why would you pay $8 billion? You're paying 3 years ahead of time for what would be a public-company multiple. Jason, you say do the deal. I don't understand. I'm really naive here. I'm not arguing; I'm trying to understand. You're paying 3 years ahead of time for what that would be in a public market.

Jason Lemkin

Well, look, if we go from 150 to 450 next year, then we're paying 20X revenue at the end of next year. So we're paying a year ahead if it triples next year.

Rory O'Driscoll

If they can peg the growth, we're all saying the same thing in different ways. If the growth stays the same, at 3X even, then you earn your way out very quickly. If it declines even to 2X, which is still amazing growth, then you could be a little ahead of yourself here. That's what we're saying. It's as simple as that.

We published on this 10 or 15 years ago. In SaaS, you could use the following rule of thumb with a high degree of accuracy: the growth rate for any year was roughly 85% of the prior year's growth. If it was growing at 100%, it would probably grow at 85%. If it was growing at 85%, it would probably grow in the high 60s. It was just a rough rule of thumb. It was roughly right.

The thing here is, instead of saying that best-in-class growth is 3X, you're seeing best-in-class growth at 10X. As I say, if that only declines slowly, if it goes 10X to 6X to 3X, these are amazing companies and any price can be paid. If it goes 10X to 3X to 2X, then good prices can be paid, but you could get ahead of yourself. If any of these things go down, then you're just way wrong.

It is possible to lose money on a great company, because that's the only risk you're running when you're paying $8 billion, right? There's not going to be a risk that Harvey isn't one of the greatest companies. It's a given. It's clearly the biggest. There's only 1 risk you're running. It goes back to where we started, actually, about price.

You have no operational risk, you have no product-market risk, and you have no TAM risk. The only risk you're running is you're paying $8 billion for something that might be worth $4 billion, and if it's worth $4 billion, you're wrong. So you can be wrong on price.

So far, no one has been wrong in AI. The stuff that's worked has been marked up and marked up again, and right now, that's the right bet to take. So, Jason, to your point, you go back. The short-form version of what you're saying is it has been the right bet to take so far. It could turn, and if it does, everyone will be over their skis. But so far, as you say, the person who was agonizing over his $3 billion bet and biting his nails at the start of the year is like, “Oh, my God, I'm so smart. I'm so smart. I have a 2.5X in 6 months,” because the momentum kept coming.

Jason Lemkin

I think Harry's got an important point going back to 2021. I think in 2021, we reached the peak of people paying 3 years ahead in terms of multiples. Maybe we're back to that. Maybe 3 years ahead is the limit a VC can do, even with the most optimistic assumptions. It has to tie to something when you put together the investment memo, right?

In the old days, it was 1X. Box and others, when Rory was there, their growth rounds were often at a modest discount, right? You weren't paying years ahead. You were actually paying a discount. I think it peaked around 3 years ahead, right? The classic 100X ARR deals in 2021 were really 3 years ahead.

Rory O'Driscoll

I agree. 2 to 3 years ahead.

Jason Lemkin

Yeah.

Rory O'Driscoll

Exactly, Jason, that's right.

Jason Lemkin

If Harry's asking, “Is that crazy?” time will tell. It might be. It might be peak, peak AI bubble, at least.

Rory O'Driscoll

It was crazy in 2021 because 2 bad things happened—not 1, but 2. One is you had a recession, and then on top of that, a lot of those companies became functionally obsolete in terms of where the excitement was with ChatGPT. Paying up in 2020 was a horrible decision.

I do believe in this case the big advantage you have is, as I say, the direction of travel is pretty clear here. I don't think we'll wake up in 2 years and go, “Oh, my God, we're not using AI for legal.” That's not a thing. I can see why people are leaning in, which is why I go back to my comment: the only remaining risk is some kind of TAM and competitive dynamics. Can you get enough of that market, and is it big enough?

That's why the leaders are getting this premium, because you can just say, “I'm going to be the leader. It might take time, but it's a big TAM. I got it.” We'll see.

Jason Lemkin

Yeah. Well, it might be that we're underestimating a risk that a lot of apps can be Jaspered.

Rory O'Driscoll

Agreed.

Jason Lemkin

There might be a GPT-6, GPT-7, or Anthropic 5 risk where we thought we had a stable place in AI, so of course Harvey and others are going to win, right? Because they control the customer, they can swap LLMs out.

But the Jaspers of the world and multiple others thought they had winners, and then AI changed and they became of very limited value. I don't think we should assume AI is stable, and I think that's why there's a code red at OpenAI, because Sam said it's not stable.

Rory O'Driscoll

I think that's fair. I actually saw a good piece. I can't remember the gentleman's name, and he was very skeptical about most software application companies in the age of AI. His comment was, “Models will do most, so you either help make the models,” which is building the AI, “or you have to do very different apps that were only possible in the age of AI.”

The comment he was making is exactly that: whenever you do an AI app, one of the non-negotiable disciplines before you do it is that you should go and try to use the core models to do the same thing, and you should try and build a functional version of that as simply as you can. If you can get even vaguely close to it, you need to pause.

For what it's worth, I don't think every app goes away. I don't think someone wants to maintain their own Salesforce. But I do think, Jason, you're right: one of the known risks in all these investments is model improvement. We're seeing it in coding, where the models are clearly trying to eat the apps.

Jason Lemkin

Yeah, there could be. I'll give an example. Replit was around for 8 years before Claude 4 made it work. Gamma was around for 4 years before it worked.

We feel like there's a stable plane, but it could be that as AI evolves, as we're able to do incredibly complex, long-context windows and deep reasoning in seconds instead of 15 minutes, I could imagine a new generation of founders developing legal software and other software that takes advantage of deep thinking in milliseconds, which seems impossible today. I don't want to wait forever for these slow analyses, and all of a sudden, it works like magic again. It's an order of magnitude better.

All of these apps could rotate out their LLMs, but it might not be that simple. These could all become obsolete in an era of infinite deep reasoning. They could.

Harry Stebbings

I think this actually goes back to our discussion about Sierra's implementation within enterprise, which I've seen with Solve. A lot of the defensibility, the moat, the skill is in the GTM and the implementation.

And so I don't think it's quite as simple as, “Oh, well, a new model will happen.” I think that's a really, respectfully, Silicon Valley view.

Keith Rabois

No, but I disagree. I think that if you could take a Sierra example with Macy's, or whoever else the customer is, ingest all their data, do massive deep reasoning with a different model, and solve customer problems in milliseconds—none of these products work in milliseconds, and they all still have some set of issues around them, right? People would immediately move to a vendor that could be 10 times better than all of them. Sierra, Finn, Schmidt, Nguyen—they are 10 times better than what we had 2 years ago.

If you've used these products in the field, you can see that, on the one hand, they're amazing. On the other hand, they're slow. On the other hand, they're complicated to set up and ingest data. On the other hand, hallucinations are a minor issue, but they're not a non-issue. They're not a non-issue, especially when the answer has to be correct. If, as investors, we have to imagine a world where that all goes away in the next 2 years, okay, it's not stable, and all of our investments could go to 0.

Harry Stebbings

And if I'm pushing back on you, I'm saying Wilson Sonsini, Cooley, the biggest law firms in the world, have all been piloting and building relationships, building implementation pathways, and integrating them into their workflows for the last year. And just because Anthropic comes out with a new model that is 3% to 5% better, they're going to switch it out.

Jason Lemkin

I'm not talking about that. You're not listening to me. I'm not saying 3% to 5% better. I'm talking about a step function.

Harry Stebbings

Sure, but we benefit from that.

Keith Rabois

How much time do you spend coding with deep reasoning? How much time?

Harry Stebbings

But we benefit from that.

Keith Rabois

None. You don't even know what you're talking about here, Harry. I know what I'm talking about.

Harry Stebbings

That's complete crap.

Rory O'Driscoll

Whoa, you guys.

Harry Stebbings

It's complete crap. That's complete crap.

Keith Rabois

You—

Harry Stebbings

We benefit—

Keith Rabois

You have a hubristic VC argument that you think because—

Harry Stebbings

Not until we know—

Keith Rabois

Wilson Sonsini invested in Harvey—

Harry Stebbings

But they're not. They're bullshit.

Keith Rabois

They're not going to switch in 5 years.

Harry Stebbings

Bullshit. We sit on top of them. Harvey sits on top of it. It benefits from it getting better. Okay? It gets better.

Keith Rabois

That is correct. Literally today, I'll tell you—today, for example, this is not even what I'm talking about with deep reasoning, right? But literally today, Replit changed. I don't want to overtalk about Replit; I just spend so much time in it, okay? Now with Replit, you don't pick your model anymore.

Harry Stebbings

Right.

Keith Rabois

It's all dynamic. It's all gone. You don't pick a high-reasoning model, you don't pick short reasoning, you don't pick anything. It just rotates through Google and others. That, in a way, plays to your point, but it's so much change.

Harry Stebbings

No, but that plays to my point.

Keith Rabois

You're right. But I don't think you use the deep reasoning enough. I mean, 15 minutes, 10 minutes to get an answer? I'll tell you, let's find a Sierra implementation, because I've tried with Decagon and all the rest. They make plenty of mistakes today. So does Fin, and these are great products. They're not 100% right.

Rory O'Driscoll

I do believe that the domain focus, the workflows, and the customer commands give you some kind of lock-in, and if the models move slowly or you stay on top of them, I think the apps companies should be fine. But I do think it's fair that if you snooze, you can be displaced.

I can imagine if, instead of the next generation of either Anthropic or OpenAI being another incremental turn of the crank, there was some kind of reasoning step function—which I'm not holding my breath for, by the way—or an AGI step function. I can't even say that without snickering. You could imagine another turn of the crank that made everything you've done so far no longer be the way you do it.

As long as that risk is there, at the bare minimum, I haven't seen any of my companies at the apps level totally go, “Oh my God, we used to do this. Now we don't.” They haven't been, to use your phrase, Jasper, right? But I have seen a lot of them have to wrestle with their offering quite significantly because some things you thought you could do are uneconomic, and then other things that you couldn't do 6 months ago are now way more compelling.

What I would say is, product-market fit is a rolling feast at the moment, and it's a moving target. Which is why, if you look at all your good apps companies, one of the things you see is that when the new model comes out, they're on it that afternoon. It's pizzas and late nights, and you have to know what's in the box pretty damn quickly. Because until this pace normalizes, you can be displaced.

Jason Lemkin

I guess here's my point. We'll just see. I'm just talking about B2B apps. I don't think—

Harry, maybe I wasn't being clear, and I was making 2 seemingly conflicting points, or making 2 points at the same time. I think we haven't even begun to see deep reasoning in B2B apps. We haven't even started. We haven't even started. We are doing relatively simple prompts that can get answers back in a second or 2 or less, okay?

You can't wait 5 minutes for Sierra or Decagon or Fin to give you an answer. No one's going to wait 5 minutes for that answer, okay? And this is part of the code red at OpenAI: they want to move away from reasoning. But I think when that goes from 5 minutes to 5 seconds, or better yet, 1 second, just like we thought Jasper was amazing, we haven't even begun to imagine what B2B will do to AI when it can do deep reasoning in 1 second.

It's just not accessible to B2B apps today. We just can't wait 5 minutes for Fin to tell us an answer. I think we just can't wait. We don't have that time. I need to find out what happened with my sweater.

7. The LLM Switching Problem

Harry Stebbings

It's what I find so funny when people compare the LLM market to the cloud market, though, because the thing that you don't have in the cloud market is the promiscuity that you do in the LLM market. I've spoken to 2 companies today, and they're like, “The upgrades in Anthropic recently have meant a full shift to Anthropic away from OpenAI and improvements in our product that we really didn't expect, which is amazing. We're so grateful to Anthropic.” But it's a real shift overnight for them.

Mark Suster

Agreed. And actually, that segues to—clearly, you're not capable of keeping us on track anymore, Harry, so I'll just cover the agenda from my side. You were going to ask us about the Benioff comment on LLMs being a commodity, which is exactly where you're going here, and it's a super interesting comment.

Let's start with the point you made: I've been struggling to find, because you always try to think in analogies as an investor, how sticky, commoditized, and profitable the models are. These are all related terms, but not the same. In cloud land, there was an oligopoly of Amazon, Microsoft, and Google, but you couldn't easily switch between them. People muttered clichés about how, “I want to be multi-cloud,” but no one's really multi-cloud, and many of my companies move from cloud A to cloud B. It's 2 years, it's millions of dollars, and it's a pain in the ass.

Here, you're right. These are APIs, and most companies have multiple providers on the same system and switch between them. In fact, Sierra did an interesting piece on how a lot of their application is a constellation of models. In real time, you're always looking at multiple things. So you're right: it's way more fluid in terms of your ability to shift between than anything you've seen before. So, I do think that's true, and the question is, what does that mean?

Jason Lemkin

And of course Mark's right. Just like Replit today, Salesforce can auto-rotate models without you knowing it. Salesforce is and should be doing the same thing, and they have their own LLMs, as he talked about, right? That's great today, and I think it's an interesting topic for 2026.

Just going back—not to push—but we use Agentforce ourselves. It is great. I can tell you how it really works. It takes about 30 days to train and deploy, and then you've got to keep going and upgrade. It's great, and it works just as well as the other agents. In 1 way, it works better than all the rest. Just 1 way, and it's Captain Obvious: it is native to Salesforce data.

So it actually works better than all our other agents in Salesforce. We can show you that. Okay, so it's great. Use it. It's going to be a multibillion-dollar business. But imagine when, instead of taking a month to train and then a couple of other months to iterate, it can happen in minutes. This will be nothing like the AI we have today. We're just starting.

Not only will that probably make Salesforce a better company, it may disrupt all these incumbents when all of this can be done instantly—when the Harvey that we used in 2025 seems just quaint in 2 years. What a joke that we had to spend all this time training it and ingesting it, with all the mistakes it made and the associates. And so I don't know. I think it's very exciting, but I think it makes our investments even less stable. This is going to change.

Harry Stebbings

And it's going to be very exciting to go from taking months to train an agent to taking minutes. It's going to come.

Rory O'Driscoll

I think the interesting question on this, from Mark's comment, was that there was an implication there that because they're, quote, “swappable”—which is true—the LLMs themselves are bad businesses, right? That was the implied statement.

Jason Lemkin

That was the diss.

Rory O'Driscoll

That was the diss. I read that as the implied statement. It's interesting to say why that's not the case, because I don't think it is. I think Mark actually used the analogy that it's like hard disk drives, and if you read Clayton Christensen, it's the prototypical example of a wildly competitive, commoditized market where there were 30 providers.

I think the analogy breaks down because, if you look at what happened in the hard disk market, capitalism works. If you have 30 commodity providers and you don't invoke antitrust, 27 of them go bust. You end up with 2 or 3 providers—the survivors—and that's what it takes so they can extract enough profits to survive. Thirty people don't get to make it.

I think if you look at the hard disk market, it was brutally competitive for a while. It remained a commodity even after it consolidated to 3, but they were able to extract enough profits from it. So I think Mark's sentence—“Oh my God, this is a commodity market like the hard disk”—might be true at the start, but there's an embedded comment there that isn't true.

Rory O'Driscoll

I think it's totally plausible that even though there's a fair degree of switching possible between models, you still will evolve to an industry structure where those model companies make decent money. Noah Smith had a good piece on this, even though he was negative on the comment. He said the model companies could be a little like the airlines: super-high fixed costs, easy to switch, and his implied comment was that that's a bad business. It's not an awful business if you consolidate.

If the antitrust people would let them, we'd only have United in San Francisco, and then they'd just charge through the nose. So I think you would see consolidation.

Harry Stebbings

How do you think about the volatility of revenues that's inherent in a business where the quality drives so much of the switching? Anthropic will suddenly have mass usage that drives revenues, and then others will lose it.

Rory O'Driscoll

It's a good one. I think what happens is businesses evolve. If you're going to be a surviving business, you can't live like that in the long term. It's just too hard, so microeconomics will make it happen.

I think it's probably 2 things. One is that as the cost to enter gets higher—the fixed-cost barriers—I don't know if there are going to be 10 of these providers. There might be 3 or 4, but I doubt there'll be 10 if it really does cost a lot to make these. If it doesn't cost a lot to make these models, then all bets are off.

Then, obviously, the second way—and you're seeing it right now—is that if, at the API level, I'm a commodity and there's me, OpenAI, and Gemini, and they're swapping between us, I've got to grab the apps. That's what you're seeing. Anthropic is grabbing the coding app, and obviously OpenAI is grabbing the ChatGPT consumer app, right?

You're basically saying, “I can't be sitting in the back here getting swapped out by some rotational round robin on AI. I've got to own the end user here.” I think Marc is right that it's a competitive industry, but I think the—

Harry Stebbings

I don't see the volatility of revenue on the consumer side because you have memory. I'm seeing this so much in the LLMs that I'm using, where memory drives so much of the actual end product that I get, and that's incredibly sticky and will drive a lot of value. But on the B2B side, where you can have companies switch so easily, there isn't that stickiness.

Jason Lemkin

Well, just 2 thoughts. On the B2B side, we talk about a couple of LLMs, but so many of the hot companies use their own LLMs and Chinese models. Cursor's own model is a Chinese model. Windsurf's is a Chinese model.

All of our portfolio companies that are using their own LLMs aren't really using their own LLMs. They're using cheap open-source models, or, if they could, they would use Chinese-hosted models, and sometimes they do because they're a fraction of the cost. So where exactly this all goes, we don't know, but all of these startups that have their own model are using Chinese open-source models. They all are, including Cursor, right?

So there's a whole other group that's competitive because they're not raising massive venture capital, and maybe they're inferior in some ways. But if Cursor's using its own model 30% to 40% of the time, it's not that inferior, because Cursor's doing pretty good.

Comment number 1: Chinese models are already being used by our hot startups. It is a real threat. Idea number 2: Harry, you think memory is a moat and permanent, and it absolutely is.

But I think the reason Sam called it code red is—imagine today, when we're doing this, if ChatGPT went away tomorrow. Okay, here's what's changed. Google AI Mode's pretty good. Now, it's not going to be our therapist, and it's not going to have as much memory, but AI Mode's pretty good.

The Google stack's pretty good, and if OpenAI, for some reason—if ChatGPT went away tomorrow—we would survive. We would complain. Keith would have some good comments. You'd have some comments. Twitter would explode, and in a week we'd be like, “Whatever, we've moved on.”

Some folks would go to Claude, even though we've forgotten about Claude, but Google's pretty good. This is why I think Sam called it code red, because OpenAI is a consumer company, and he even said, “Listen, we're going to do even more on ChatGPT. We're going to do less on apps, we're going to do less on imaging and video. We've got to make it good.”

We would be in a tough spot for a while if Anthropic went away, because all our apps wouldn't work. That would suck. But if ChatGPT went away tomorrow, we'd be fine in a couple of days. We'd lose a little memory. We'd have to rebuild it in Google, but we use Google every day.

It would not be the end of the world. It really would not be the end of the world if this massive thing, which is the greatest consumer app, went away. It would not be the end of the world.

Rory O'Driscoll

I don't know if I buy that. It's not the end of the world, because nothing is the end of the world unless it's the end of the world.

Jason Lemkin

It'd be worse to lose Netflix. That has more proprietary use cases than ChatGPT.

Rory O'Driscoll

That's a more interesting comment, actually—genuinely. It's like I'd pay $20 a month—

Harry Stebbings

It's not not true.

Rory O'Driscoll

Could you not apply that to all of them, including Anthropic? I'm genuinely asking here, Jason. You're the coder here, and I submit to you on this one. You could switch Anthropic out, and it would get 90% efficiency on other models. Netflix, like, it would just be like, shh.

Jason Lemkin

Well, there aren't that many.

Keith Rabois

But—

Jason Calacanis

I do think Anthropic is more enterprise. In my experience, it is stickier because of that. There's more work than just moving in a day. That's why I slightly challenge your point that you're saying the consumer was stickier.

I do think memory's a big deal, right? It's getting really, really, really good, at the edge of creepy. But it'd be like a TV show. We would just shrug, and we would switch the next day. The greatest consumer app of all time so far—we'd be fine.

Rory O'Driscoll

Put me down for not agreeing.

Harry Stebbings

Yeah.

Rory O'Driscoll

I think we would—look, obviously we'd get on with life, because 3 years ago, apparently, we were without this and we were fine. But I think it would leave a significant gap. There's no doubt that it's a superior product and you'd miss it. So I disagree.

I think the test you'd use is this. When we were looking at consumer hardware once, one of my colleagues gave me a mental test that I actually think was one of the most insightful things I heard about consumer products ever. He said, “The only test you need to run on this one is this: If you lost it today, would you go out and buy another one tomorrow morning?”

If you lose your phone, you instantly go out and get one in the morning. If I lost our coffee maker at home, I would be down to Solo Travel in an hour. Then there are other things where you go, “Yeah, would I take a week? I might get round to it. Maybe I wouldn't bother,” right? And it's just a really good test.

Using that test for ChatGPT, I think if I lost ChatGPT now, I'd be like, “I've got to go find something,” and I don't think Gemini's quite there. So I think it would leave a void, which speaks to their strength, because maybe it's not an “I would die” void, but it's 800 million people who use it.

I think there's a lot of gravity in those individual consumer users and the habit involved. I think it's their biggest asset.

Jason Lemkin

But it could be just as sticky as Yahoo Mail. We survived and moved to Gmail. It was okay.

Rory O'Driscoll

It was, but it took a sustained 5 years of under-execution by Yahoo to get that done, and I think Yahoo Mail and Gmail were more similar, I would argue. Even today, despite all the progress in Gemini, I think the ChatGPT experience is still better. Controversial take, perhaps.

Jason Lemkin

Play on it some more. Let's talk about it in January. I think you're going to come around to my position, which is just the same one that Sam has: it's pretty good now.

Rory O'Driscoll

I don't think it's going to stop OpenAI's future, but everything's just unstable, including Harvey, I think.

Harry Stebbings

I think what you're seeing with OpenAI is the awareness that they've been massively diversified, way too much. They need to consolidate. They need to concentrate. They are absolutely aware that they have fumbled the ball over the last 12 months, and they need to massively reduce the portfolio of products that they have.

There's a focus on healthcare, there's a focus on codecs, and there's a focus on, obviously, the consumer product, and that is the BFD for them. I think Brad feels the pressure, I think Sam feels the pressure, and it's game time.

David Friedberg

Yeah.

Harry Stebbings

Rory O'Driscoll

Yeah. I just think the way markets work is there's this formative period where there's a lot of change, and then it locks in, and then change is much harder to happen. Remember, we were talking to Tomasz, and he had some congealed grease thing.

My mental model is always that it's like the Big Bang. Early on in the Big Bang, everything happens, and from then on, it's just expansion and nothing changes, right? This is one of those moments in time where how you end up ranking by the end of '26, '27, probably, I think, will determine the trajectory for 10 years.

I think the variance, the high fluctuations we're seeing now, will steady out. I can't prove that. You could have a world where it remains constantly changing, but in my gut, economics tends to force that kind of stability. So you really want to be sure that when things start to lock down, you lock down in the number 1 position.

8. The Chinese Model Debate

Jason Lemkin

We'll move on, but Andreessen said a couple months ago that 80% of its startups use the Chinese open-source models for AI.

Rory O'Driscoll

That's not a correct quote. My understanding—Martin Casado then tweeted and said, "That's not quite a correct quote." I think it was something like, "Only 20% of the time, my companies are using open source, and when they are using open source, 20% of that open source is Chinese open source," which is a very different thing, but still a significant thing.

It's an interesting comment here. What it says is that there is a demand, no surprise, for a quote-unquote cheap open-source model. Given that both formerly idealistic, now highly commercial corporate entities have stopped providing state-of-the-art open-source models, there is a latent demand for that which, for their own good reasons, which we can speculate about in a second, the Chinese tech industry appears willing to fill.

Even though the market tells them to stop doing it, it was kind of nice when Zuckerberg was willing to fund R&D development and keep churning out excellent Llamas for everyone to leverage.

Harry Stebbings

Jason, should US startups be able to leverage Chinese models?

Jason Lemkin

Listen, assuming there are no security issues, I don't care. I don't fully understand inference because, in most cases, if you're running a Chinese open-source model on Amazon, it's really not any cheaper. You're not saving that much. You save a little bit of money. Correct, you don't have to pay for the model, but inference is where most of the cost is.

I think there's still this gray area where, obviously, it works for Cursor at scale. Exactly where it works, I don't know. But look, as long as we're not sending data back—and maybe we shouldn't—I couldn't care less.

More importantly, listen, I just want to make money like everyone. It's 2025, going into 2026, Harry. Don't tell me about it as an investor. This is an area of unlimited greed. I don't want to hear about it. We're going to forget about it in a week anyway. We already forgot about Real and Dipling. We forgot about all of it. I just want to make money and be quiet, okay? Just don't talk about it. I just want to make money.

Rory O'Driscoll

The question on, quote, Harry's, "Should we use Chinese open-source models?"—you have to break it into 2. Are there national security issues or regulatory issues? It's the job of the government to figure that out, and we should take our lead from that. If the government says no, you should abide by the law. If the government says yes, you should at least take that into account, and you can say there aren't national security issues.

Separately from that, you have the business issues. Is it good value? Is it cost-effective and all that stuff? I mean, I think the interesting thing is because, Harry, I've watched your little dispute with Keith on Airwallex on Twitter. The odd thing is, I can't quite figure out what I'm—

The national security guidance on China is quite confusing, because obviously we're now back to selling them advanced NVIDIA processors. I think the logic is some version of, "They're going to figure it out anyway, so sell to them." So it's not clear to me where we can say we can do it and where we cannot. Separately, if you can do it legally, then after that it's just practical and functional: if the models work and you can do it, you can.

Harry Stebbings

So I'm getting into a very dangerous zone because Keith Rabois is a friend and an LP of mine, and Airwallex is a $7 million investment of mine. So I'm screwed both ways on whoever wins and loses in this one.

9. Airwallex Faces China Scrutiny

Airwallex raises $330 million at an $8 billion valuation, led by Lee Fixel at Addition, a big investor in Stripe. Then, in the last few days, Keith Rabois has been very public in stating that there is a data concern about the data flows that will go back from Airwallex to potential Chinese officials, government, you name it, whatever institution that is. That is the news.

Jason Lemkin

Based on the conversations we had, $1 billion in ARR, $8 billion valuation. Why is it so cheap?

Rory O'Driscoll

It's because it's an excellent, excellent payments company, and it'll be valued like a payments company. I mean, excellent growth rate, excellent payments company, with an Asia discount, right?

In other words, it's cheap in the context of not having the AI magic pixie dust, and on top of that, you have the Asia discount. You're exactly right.

Harry Stebbings

No, but yeah, Rory, it's doing the same revenue as Ramp does. It has more integrations and more underlying infrastructure. It is absolutely the Asia discount.

Jason Lemkin

Yeah, let's assume it's growing the same as Ramp and has the same gross margins, right? It's the same ARR. Maybe the margins are different, but let's assume it's the same for the sake of argument. It's trading at a quarter of the price.

Harry Stebbings

It's the Asia discount.

Jason Lemkin

I get why it's dual-headquartered in SF. It just seems a quarter of the price.

Harry Stebbings

I completely agree with you, which is kind of why it's such an obviously good deal.

Rory O'Driscoll

I'm not sure. I think a significant portion of it is the Asia discount. I also think—

Harry Stebbings

What is it if it's not that, Rory? I would love your view.

Rory O'Driscoll

One is, I think, that versus the median fintech, Ramp is the other extreme. Ramp has an extraordinarily good revenue multiple. My wider comment is that fintech multiples tend to be significantly more bounded than open-ended AI multiples. Ramp, you're right, is the most extreme.

Harry Stebbings

No, but I would actually push back and say, look at Brax at $13 or $14 billion with a much worse growth rate, much less profitable, and $400 million less in ARR.

Rory O'Driscoll

I would agree with that. That difference between that median one and where you come out, that is some kind of Asia discount. You're exactly right.

My point is that between that and Ramp, I would argue—because, look, there's obviously a difference between Ramp and Brex, and I think that's the high-end execution, et cetera, et cetera, market. So there are 2 things in the discount, is all I'm saying.

But yes, let's stipulate, to advance the conversation, that there's a significant Asia discount in here.

Jason Calacanis

75% off is a lot.

Harry Stebbings

Yeah, it's a huge amount. The question for me becomes, actually, when is that shed? Because in the case of ByteDance, it's never shed. ByteDance is still massively discounted versus a Matter, despite it being a phenomenal business both in margin and revenue scale.

Rory O'Driscoll

And that's a good question. First of all, I think the ByteDance situation is trickier, because, one, ByteDance itself is clearly a Chinese company. Then what you're really talking about is, are you talking about the ByteDance aspect, or are you talking about the fact that you have the whole TikTok dynamic?

Because that's a media company present in the phones of millions of Americans that has an end owner that is ultimately based in China. So that's that discount.

Jason Calacanis

Yeah.

Harry Stebbings

Yeah.

Rory O'Driscoll

Well, there should be a China discount, not just because of us. For all the problems with our government, the Chinese government—the dictator of China—said about 5 years ago, "You shouldn't be doing social media. You should be making hardware shit." And he has complete and untrammeled power.

So there's a reason there's a discount. The reason there's a discount is the one-party dictatorship that runs the country doesn't think those are good businesses. So it's not like it's a China discount for nothing. Now the question is, is it too much?

Harry Stebbings

Can we just appreciate—sorry.

Airwallex is not a Chinese business.

Rory O'Driscoll

I know, but you mentioned ByteDance. I know.

Harry Stebbings

No, I'm intrigued. Zoom has many more engineers than Airwallex.

Rory O'Driscoll

So you mentioned ByteDance, so I went down that rabbit hole. What you're really talking about in the context of Airwallex and many other companies is something like the following. These are absolutely not companies headquartered in China or owned by Chinese investors. They are U.S. or international companies. International in the case of Airwallex, I think they're originally based in Australia, headquartered in Australia, subject to Australian law, et cetera, et cetera.

So when someone says that something is a Chinese company, that's grossly incorrect. What is true, however, is that if you have a significant number of engineers or data centers based in China, then you are also subject to Chinese law, just as when you have data centers in Europe, you get the Europeans on your case. And if you're a European company based in America, you get the Americans on your case. All the superpowers—if you lump Europe kindly into that—any of those countries, or entities in the case of Europe, regardless of where a company is headquartered, have a regulatory role to the extent you're operating in those countries.

Harry Stebbings

So my legitimate question back to you is, if Zoom has more engineers there than Airwallex does, why are we not holding them to the same account? Microsoft and Apple do too.

Rory O'Driscoll

It's a great question, and I think every one of those, to the extent that these kinds of issues start to crop up, is going to be a thing. Once you start doubting the other side's bona fides, things just get worse. And it's probably true to say that if you have a critical mass of engineers, and in particular if you have a service center based in China, you should assume, just under standard Chinese law, that government entities have the right of inspection. Just as, in a similar, not quite the same way, when you're based over here, you have all the rights of inspection the U.S. government would have and access to information.

I think we're more of a rule-of-law country, but I still think all 3 of the big entities—including, as I say, Europe as the other—have some version of that. So if you choose to put your back-end data center or back-end employees in China, you're taking on that risk, and you're vulnerable to this kind of statement where someone says, “Oh my God, your key information is flowing through a Chinese data center, flowing through Chinese employees. Maybe it shouldn't be.” The further apart the countries grow, the more this kind of thing gets problematic. If you were trying to sell one of these companies, the embedded risk in China would be a key consideration. A U.S.-domiciled company buying one of these companies would be thinking about that.

Jason Lemkin

You probably will struggle to get some U.S. government contracts. It's just going to be a thing more and more. So I think you're just going to see a gradual separation. You'll just examine the risks and go, “Yeah, I can get great engineers. I can save 20% or 30%, but in return I get embroiled in arguments on Twitter, but more importantly, I probably get precluded from government contracts. I have an extra layer of due diligence. It's just too hard.”

Just like all those Chinese companies that went public in the U.S., right? Let me give you the counterexample, because we all see it from our perspective, and I do think we're the good guys. But it's also interesting to step into the other party's shoes. A lot of Chinese companies went public in the U.S., and the SEC said, “As part of being public in the U.S., we want to have access to your audit papers, including the audit papers for your Chinese auditors.” They said no, and as a result of that, we pushed a lot of them off being listed in the U.S.

You're going to see this kind of thing happening more and more. I mean, look, we're not even friends with Europe anymore after our national security policy, so you're probably going to see some of the same thing. We're dealing with a gradual unwinding of globalization, and this is what it looks like. And to your point about you liking no one except Keith yelling at you, the competitors are going to be the people who use it first, because if I was an all-American fintech company, I'd have my all-American flag out there and say, “These guys are a bunch of commies. They have commies looking at your data. Our stuff is in Texas. We win.” That's going to happen.

Harry Stebbings

I think the thing I would say to Jason's point about “no one cares”—no one cares. You know what? I know Alex very well at Deel. I'm not commenting on what happened there, but zero churn. I know Jack and I know Airwallex. Zero churn.

Jason Lemkin

Well, I suspect—listen, a couple of things. First of all, I suspect this won't lead to churn. I think this is probably an offensive aimed at denting progress, right? Let's be clear: I believe you that it helped. I do think the goal is to impede, whether it worked or not. I would just say 2 things.

1, I don't know the CEO. He seems ultra-impressive from afar, not just from the recent press, but from following him for a while on social media. When this happens, you have to not only say, “No, it's U.S. data sent to China”; you have to make it unimpeachable. You just have to remove the objection. An objection has been added to a sale.

Let's just step back for a minute. An objection has been added to a sales process, maybe to derail a financing, maybe to limit the size of the round. Who knows? It's probably linked to the timing of the round. It's an objection. Is there maybe a racist element to it that I don't like? It feels like it. It feels racist to me. I don't like it. But my point is, from a sales perspective, it's an objection that's going to come up a lot, right? Maybe Ramp has a security issue next week they have to deal with, for example. You have to deal with it.

If I know my data isn't going to China, 100%, I think outside of the U.S. government, the objection's over. I do think, to Rory's point, in my experience, you're never going to get a government contract. But so what? The world's big. You're just never going to get a government contract. So what? That's just life.

Rory O'Driscoll

Oh my God—a sentence I never thought I'd say: defending Keith Rabois. I don't think it's racism. Look, I genuinely don't.

Jason Lemkin

I don't think Keith is racist, but I think—

Rory O'Driscoll

Yeah.

Jason Lemkin

There could be an element of racism here. It's very easy to pick on someone that, my understanding is, is not Chinese, is Chinese Australian, and they did it with Eric Yuan at Zoom. It was very racist for a while. It was very racist when people wanted to make it a Chinese company, and Harry's right. They do have a large Chinese presence. It was a vulnerability for Zoom, right?

Rory O'Driscoll

I'll disagree. I don't want to devolve down to it. I think there are legitimate security questions when you read the laws of the relevant countries. Look, let's give another example. When the Europeans get all bent out of shape about our lack of privacy laws and put up their own weird laws to stop us because they're worried about us, that's just a mirror image of this.

I don't think this phenomenon is driven by anything other than large country blocs with very different perspectives increasingly being uncomfortable with how the other countries or systems are governed and wanting to exert power in a world where they can. It was great when we were the dominant power. Everybody did what we wanted, and it was freaking awesome. Everybody did what we wanted because they wanted to be in our game. Now, for whatever reason, we've convinced ourselves that that was bad, and now we're going to have a world where people don't want to do what we want. Then we're going to find it's a bigger pain in the ass.

10. The Final Investment Picks

Harry Stebbings

All right, boys, this has been fantastic. I'm going to push you to 3 companies, okay? And you've got to pick 1.

Rory O'Driscoll

Oh.

Harry Stebbings

Yeah, I know. Rory, you love it. I almost do it because I know that you really can't—

Rory O'Driscoll

I know. You do it to humiliate me.

Harry Stebbings

No, not to humiliate you. To force great thinking. Airwallex at $8 billion or Ramp at $32 billion? Which would you rather own?

Rory O'Driscoll

Jason, bail me out here. I'll tell you: let's assume the revenue and growth rates are the same, and I don't think they are. I don't have the time, but I think Ramp is a more enterprise-y product, and Airwallex is more about money movement. I think that's a significant point.

But just to give you an interesting discussion versus a boring one, if I was on the Airwallex board, because I don't think it's, quote-unquote, a Chinese company—I think it's a perfectly normal Australian company—I would put in an $8 billion valuation, and part of my closing condition would be that within 24 hours, we do not have a single paid employee in China, and we relocate the entire service organization to some other jurisdiction.

Then I would do it at $8 billion if they are equivalent, because I think that's something you can do. I'm going to give them credit. I think Magnus did a great job on that. I think Benchmark did a good job of making it very much a company of people who used to live in China. That would be a value-add moment.

I've had that with another company where, I'm not going to mention names, we had a significant Russia presence, and this was before 2014. I was like, “Let's just hire people somewhere else. It's not worth the extra 20%.” So I think the same thing applies here.

I would do Airwallex at 8 or 9. I would go on that board and say, “Figure out your location strategy and get relocated out of China within the next 12 months.”

Jason Lemkin

I wouldn’t make that a condition at all. I would just trust Jack to figure it out. Obviously, he’s well aware of the issues and the trade-offs. He’s well aware engineers are global.

I think a VC condescendingly telling the CEO how they’re going to work it out is beyond condescending. The great CEOs will figure it out.

Rory O'Driscoll

I’m going to push back hard on that. Hard on that.

Jason Lemkin

Okay. But people are going to see you as condescending, Rory. I wouldn’t push back.

Rory O'Driscoll

They like you a lot now. Don’t make them think of you as condescending. It doesn’t help at the end of the year.

No, I’ll deal with that straight away.

There are a few things where it is a board’s place to say something, and this is one of them, where you’re like, “Strategic fatal-error risk is exactly what a board has to do.” The ability to sell the company is something a board has to work on. I remember this other case: saying to a CEO, “Look, you will limit your pool of buyers dramatically if you have this significant exposure to this thing.”

You can talk about how you get rid of that as a CEO, but it is one of the few things where you, as a board member, might have a little more perspective than someone. When you’re down in it, you can go, “That’s silly. It’s wrong. I know these people. They’re trustworthy. It’s not really an issue. They don’t have access to the data.”

One of the roles of the board member is to say, “Step back, dude. Everything you’re saying is true, but when it goes to shit, no one will care. Why pay the tax? How much extra would it cost?” So I disagree. That would be one where I would think that would be something you should say. But this is good, because we’re meant to have disagreements.

Jason Lemkin

I think you just do the deal at Airwallex and trust them to figure it out. I think it’s that simple. 8x ARR is a good deal. You trust great founders to figure it out. I think it’s a fine deal. It sounds like a dislocation in the market. You could do worse than invest in a dislocation in the market.

Harry Stebbings

Kalshi raised 1 billion at 11 billion.

Jason Lemkin

That doesn’t seem like a dislocation in the market. That seems like a location in the market—a very precise heat-seeking missile, just like Maverick and his team of Tomcats hit that right in the movie.

Harry Stebbings

Dude, prediction markets—it’s a thing. You can have Kalshi at 11 or you can have Polymarket at 13.

Jason Lemkin

Whichever one is better for insider trading, I’m going to do that, right? Because that’s really the exciting part, right? Either betting illegally on sports or, even better, betting legally on confidential Google and Meta information like they’re doing, right?

You don’t have to be Nancy Pelosi to make money legally out of confidential information anymore. You could just be a senior engineer at Meta or Google and make millions on the side. It’s a great time to be alive.

Harry Stebbings

He’s come out punching today, hasn’t he, Rory?

Rory O'Driscoll

And just in case anyone doesn’t know this, these comments are driven by an anonymous prediction-market person who had made literally millions of dollars betting on things like what the number-one query term on Google would be that day, and getting it right for sequential days in a row, leading the suspicious mind to assume that they have, in fact, access to that data. Correct, Jason? You’re exactly right.

Jason Lemkin

Well, also, Pony Ma’s bet on OpenAI news. He’s doing great. Just being able to sell 20 million every 2 years isn’t enough. Now he can day trade on the information. These are great days.

Rory O’Driscoll

No, I think you’re exactly right. It is quite a phenomenon, and it will probably result in some level of regulation, because you have both insider trading on this kind of arcane information, and you’re beginning to increasingly see concerns on the sports-gambling side, especially when you can do these micro-bets.

It’s one thing to say, “Am I going to throw the game for my bet on the side?” But if you can actually bet that in the third quarter, an athlete will only score 1 basket or whatever, you can have these micro-bets that don’t change the thing. You really are very vulnerable to a person betting on something that they can control either themselves or through someone else.

So I think they’re super interesting bets. God, I wish I was in one because it’s interesting. But there’s a cesspit of issues coming here, and if there’s ever a turn to regulation, in about 3 or 4 years, just like the quiz shows in the ’50s, there are going to be a bunch of congressional hearings where someone is going to be asking the Kalshi and Polymarket CEOs, “How did this guy get that right 27 days in a row? Do you know your customer?”

And he’s going to go, “It’s all crypto-based. I don’t know.” And they’re going to say, “Hmm, maybe that’s not how we’re going to roll anymore.”

Jason Lemkin

It’s good times.

Harry Stebbings

And that’s where we have Pakistan play cricket.

Rory O’Driscoll

And by the way—yeah, someone remind me—how are you guys doing in the Ashes at the moment?

20VC OGs: SpaceX Valued at $800BN & Harvey Raises $160M at an $8BN Price | Airwallex Raises $330M and The Battle with Keith Rabois | Netflix Acquires Warner Brothers | IPO Market Predictions for 2026: Anthropic, Stripe, Databricks and SpaceX | BidClub