The Numbers Behind Anthropic's IPO, AI Bubble Questions, Meta's Spending Continues
- Anthropic’s strong operating business does not remove the cost of staying at the frontier. Thompson says subscriptions are profitable and the API is more profitable, but training-related R&D exceeds serving costs and increased more during the first half despite exploding usage. Competition determines spending whose pace is “out of their control.” He expects a bubble and painful reset, while explicitly declining to identify this IPO as the top.
- The headline loss and headline margins require different accounting adjustments. Thompson attributes the reported $80 billion first-half 2026 loss to repricing Amazon’s loan liability, which could be repaid in shares: “It’s not money out the door.” Conversely, about half of revenue comes through cloud platforms, whose roughly 30% cut—with estimates around 37% also cited—is classified as sales and marketing. He thinks that treatment overstates gross margins, although the underlying business still looks excellent.
- The hyperscalers could capture more of AI’s upside by financing the industry’s funding gap. Google, Amazon and Microsoft have the greatest capacity across free cash flow, debt and equity: the “girders of the bridge” until customer revenue catches spending. Existing shareholders fund equity issuance through dilution, and distressed labs could eventually accept worse terms. The hyperscalers that help finance the bridge could also gain power to develop competing chips and press Nvidia’s prices.
- The demand question is about timing, even for an AI believer. Craig Moffett’s test is whether final customer payments can support the entire value chain in 2027–2028, excluding transactions between suppliers. Thompson says current Anthropic revenue comes nowhere close to industry capex, but that infrastructure serves demand two to five years ahead. “Multiplying humans” into new businesses could create that demand; the difficulty is that such transformations usually take longer.
- Frontier labs need customer attachment to escape competition based solely on model quality and price. Microsoft and Palantir want customers to keep models replaceable, while Thompson argues the labs eventually need to own applications, workflows and interfaces. Cloud workspaces and accumulated habits make switching harder than changing an API endpoint: “Lock in is an important part of making money.”
- Meta’s Muse strategy makes heavy training expenditure integral to the investment case. The discussion cites up to $95 billion of stock offerings and shares down roughly 10% over a week; Sharp says dilution is a reason the shares should be down. Thompson says companies like Meta may be able to get away with dilution. Meta must control Muse’s underlying model, so the coherent objection is that “Meta doesn’t actually need Muse.” He also sees an indirect threat to Netflix: AI funding pressure pushes Meta to maximize Reels engagement and advertising.
- Government power remains a constraint on private AI and communications infrastructure. Thompson’s argument is conditional: if AI becomes as powerful as Anthropic predicts, governments will assert authority because “they have the guns.” Both hosts describe the Ukraine Starlink episode as opaque; Thompson guesses Musk acted in alignment with Washington, while Sharp guesses Washington requested the action. Neither presents those guesses as established facts.
1. Anthropic’s profitable products sit inside an expensive race
Asked whether Anthropic’s IPO could mark AI’s peak as Coinbase’s listing did for crypto, Thompson answers, “I don’t know.” He considers a bubble, painful collapse and eventual recovery inevitable, but cannot identify the timing. His concern is investment arriving faster than useful ideas can be absorbed, rather than a belief that the technology lacks value.
Sharp’s reading of the S-1 is that Anthropic looks like “a rock-solid business.” Thompson agrees about the operating core: subscriptions are profitable, and the API is more profitable still. That distinction matters because an attractive business serving existing models can coexist with a much harder economic proposition in developing their successors.
Sharp notes that subscriptions also create distribution advantages. Having already paid, customers use Claude, develop habits and advocate for it inside their employers. Thompson thinks paying for both competing services is worthwhile, but doubts many customers will. The subscription therefore purchases recurring exposure to the product as well as revenue.
The catch is R&D: training spending is larger than serving costs and increased more during the first half even as usage exploded. Better competing models force further investment. “The speed with which they need to spend money on that is out of their control.” Sharp emphasizes that this is an ongoing expense, rather than a problem solved once.
2. Cloud distribution improves access and complicates margins
Amazon’s early investment forced Anthropic onto Trainium, something Thompson doubts it initially wanted. That constraint proved useful: starting with Nvidia, then supporting Trainium, TPUs and an AMD deal required hardware flexibility. Anthropic built “a layer of abstraction” across chips. Amazon benefited too, because a demanding frontier customer improved Trainium’s capabilities and tooling for other customers.
Availability across AWS, Google’s cloud and Microsoft also helped Anthropic win enterprise customers while OpenAI was tied to Microsoft. Thompson’s concrete example is straightforward: a company whose data already lives on AWS would rather consume AI there. About 50% of Anthropic’s revenue comes through those three providers.
Thompson challenges how that revenue is recorded. If a customer pays Amazon $100, Anthropic recognizes $100, then books the platform’s roughly 30% share—about 37% in estimates cited—as sales and marketing. He argues for recognizing net revenue or treating the payment as cost of revenue. Either would reduce the headline margins under discussion; he still calls the business’s margins excellent.
3. The $80 billion loss exposes the mechanics of dilution
Thompson explains the reported $80 billion first-half 2026 loss through Amazon’s financing. Amazon supplied money as a loan that Anthropic could use for compute; Amazon could receive repayment or take shares. As fundraising rounds increased Anthropic’s valuation, its obligation was repriced and the increase booked as a loss. Thompson believes Amazon will receive shares at the IPO: “It’s not money out the door.”
The corresponding investment appreciation appears in Amazon’s income. That creates a broader earnings-reading problem: hyperscalers’ reported profits can swing with investments unrelated to their operating performance. Thompson contrasts Amazon’s Anthropic gains with a decline in Google’s SpaceX holding and advises focusing on operating income.
The economic bill survives the accounting adjustment through dilution. Thompson identifies outstanding share obligations involving Broadcom, Google and roughly two-and-a-half additional Amazon arrangements. He also describes a provision permitting issuance of 2% of outstanding shares after roughly 50% stock appreciation. “The people who are paying are existing shareholders”: their ownership funds the compute purchased with those arrangements.
Equity financing shares upside with new investors, which Thompson considers healthier than relying entirely on debt. Sharp prefers public access and disclosure to permanent private ownership. Still, Thompson questions the opportunity available to buyers at illustrative valuations of $1.5 trillion or $2 trillion. His caricature of the pitch: “We’re going to dilute you endlessly, but we’re going to control the world.”
4. The IPO snapshot may conceal a changing growth trajectory
Sharp suspects, without claiming certainty, that insufficient compute made first-half financials look better. Thompson explains that Anthropic did not have enough capacity: because compute is a major cost, the shortage kept costs lower even as customers experienced outages. Sharp jokes that those customers “did your part for Anthropic’s IPO” by helping make the margins look better.
The S-1’s numbers stop at June 30. Thompson asks whether spending indicators such as Ramp’s index showed growth slowing, given OpenAI’s push and Astra’s arrival, or whether it reaccelerated after Opus 5.5. Sharp argues that Opus 5.5 could restart growth. If model improvement is what restarts demand, however, that reinforces the need to keep funding the training race. The growth slowdown remains uncertain, and Thompson notes that extraordinary growth cannot simply repeat indefinitely.
Thompson says the financial information becomes stale on November 12, while reporting points to a mid-November listing. Whether Anthropic lists before that threshold or supplies an update could materially change the picture.
5. Owning the interface offers an escape from replaceable models
Thompson describes a second battle beyond OpenAI versus Anthropic: frontier labs versus the surrounding software industry. Microsoft and Palantir tell customers to build around models that remain interchangeable. Satya Nadella and Alex Karp emphasize flexibility, better prices and protection against handing a future competitor the details of one’s business.
That proposition is uncomfortable for labs whose largest expense rises with frontier competition. If customers can switch by changing an API, pricing power stays limited. Thompson argues that, “in the fullness of time,” labs need to move up the stack and own the customer experience. Sharp sees potentially enormous software revenue; Thompson emphasizes the resulting attachment.
Their own usage shows why interfaces matter independently of model rankings. Thompson compares Codex to people who organize email into folders: projects and threads suit structured work. Sharp, a former lawyer, calls Codex “the associate” and Claude “the partner.” Thompson says Claude handles varied incoming tasks, while Sharp has done most of the concrete building on his GeckoBot project in Codex, with Claude powering it.
Cloud workspaces deepen that relationship. A convenient virtual machine that remembers work also means, as Thompson puts it, “you’re now stuck in their cloud.” Applications and familiar workflows make switching progressively harder. That is the commercial purpose of climbing the stack: “Lock in is an important part of making money. Like, what’s your moat?”
6. Anthropic’s moral positioning raises a governance question
A listener compares Dario and Daniela Amodei’s criticism of OpenAI during their Walter Isaacson interview to “someone badmouthing their ex during their wedding vows.” The complaints span advertising, sycophancy, government dealings and honesty. Thompson’s objection is strategic as well as personal: defining decisions through opposition lets OpenAI set the agenda. “Who’s actually setting the agenda here?”
Thompson says he has never taken Anthropic’s claimed moral authority seriously. Sharp, by contrast, says he believes they are trying to act as moral arbiters. His concern is concentration: “There is no small collection of people that get to decide it for everyone else.” Thompson agrees that teaching Claude morals is not the right approach.
Thompson connects the governance concern to the S-1’s unusual structure: roughly seven share classes and special voting arrangements he describes as a “magic share.” He prefers conventional corporations with understandable profit incentives. Those incentives have problems, but give counterparties a predictable framework for decisions and investment. Public buyers receive disclosure; they also knowingly accept dilution and governance that is not simply organized around maximizing profit.
7. Final demand must eventually finance the entire chain
A listener brings back Craig Moffett’s question about 2027 and 2028: are customer payments entering Anthropic and ChatGPT sufficient to cover the whole system? Transactions among Nvidia, TSMC, clouds and labs can distract from that test. Moffett’s formulation, quoted by Sharp: “All that really matters is enough money coming into the system to support the system.”
Thompson says current Anthropic revenue comes nowhere close to aggregate capex, but that spending also builds capacity for two, three, four or five years ahead. The uncertainty is timing. Strong revenue growth makes investors tolerate dilution and unusual governance; slowing growth makes those same features suddenly salient.
Google, Amazon and Microsoft have the greatest capacity to supply free cash flow, borrow against diversified businesses and issue shares. They are the “girders of the bridge” across the funding gap. If labs require repeated rescues, the hyperscalers could acquire substantial ownership and control. Sharp imagines Amazon offering a distressed Anthropic terms even less favorable than its early financing.
Nvidia’s support for neoclouds reflects the risk that funding runs out before revenue catches up. Thompson’s railroad analogy preserves the distinction: useful infrastructure can survive a financial crash. But a successful bridge financed by hyperscalers creates another problem for Nvidia. Those customers gain negotiating power, develop competitive chips and push pricing; the entities sustaining demand can also weaken its supplier’s position.
8. Multiplying people creates demand on a slower clock
Drawing on his exchange with Palantir’s Shyam Sankar, Thompson distinguishes improving the bottom line by replacing people from improving the top line by “multiplying humans.” The latter creates workflows, opportunities and businesses that did not previously exist. It also usually takes longer and often comes from new companies. AI-native challengers may displace incumbents that struggle to bolt AI onto unsuitable software structures.
Thompson could scarcely be more bullish on AI. Sharp says he is in the middle ground and does not think AI will replace everyone. Thompson argues that the economy is not purely utilitarian: a huge portion of it involves status, pleasure and things people want but do not need. Infrastructure spending happens now; the businesses and cash flows required to repay it may arrive much later.
His defense of advertising follows the same optimism about new demand. Meta can show people things they did not know they wanted: “It’s manufacturing human desire.” Economic activity includes expensive World Series tickets, not merely necessary tasks AI can perform efficiently. Thompson calls Anthropic’s criticism of advertising evidence of misunderstanding humanity. His answer to claims that people cannot invent valuable new activities: “I bet we can.”
Thompson’s timing example is the internet: 1993 preceded search advertising around 1999–2000 and the feed in 2006. Discovering a powerful technology’s interface, businesses and monetization takes time. Sharp says AI will be good but does not think it will replace everyone; Thompson says he is an unequivocal believer in AI while emphasizing that monetization and adoption take time.
9. Muse makes Meta’s training expenditure a strategic commitment
Sharp cites Meta shares falling about 10% over a week after increased AI spending plans and a new nonvoting share class. The report discussed potential stock offerings of up to $95 billion. Sharp says dilution is a reason the shares should be down; Thompson notes that companies like Meta, Google and Microsoft may be able to get away with it.
Anthropic’s disclosure clarified the debate for Thompson by exposing training’s share of costs. Muse could become the “aggregator of aggregators,” owning the interface between users and other services. But Meta must control a sufficiently good underlying model; outsourcing that dependency to a lab seeking the same interface would be dangerous. “If you believe in Muse, you should be happy they’re spending money.”
Sharp says Muse’s model is not good enough; Thompson replies that he does not really experience that limitation. Sharp instead highlights 1Password integration as a major improvement. He explains how an agent-specific vault can supply API keys and passwords without awkward terminal files or secrets pasted into chat. Thompson sees 1Password’s unpopular transition toward cross-platform enterprise software as prescient: businesses pay, Apple pressures consumer password tools, and agents need credentials.
The training-versus-serving distinction also changes the discussion of Anthropic’s enterprise push. Sharp says the S-1 gives him some sympathy for it: enormous training bills create pressure to sell the resulting models somewhere. Thompson notes that serving enterprise customers consumes relatively little compute compared with training, easing the concern that those customers necessarily take capacity away from other products.
10. Meta’s cash engine funds ambition and pressures Netflix
Thompson says Sharp wrote the most compelling investor objection: “Meta doesn’t actually need Muse.” Its advertising business remains formidable. Investors seeking straightforward optimization would prefer the behavior of professionally managed Google or Amazon. Meta instead has a founder, Mark Zuckerberg, pursuing another platform opportunity. Sharp considers this direction more promising than some of Meta’s previous detours.
Thompson views Amazon’s infrastructure and inference focus as rational for a company already built around lower margins and logistics. It can serve Anthropic and OpenAI without spending billions to lead frontier development itself. The hosts defer their discussion of Google to the next episode.
Thompson’s Netflix thesis is indirect. AI expenses pressure Meta to produce cash, encouraging more ads and more engagement. Earlier restraint over feed ad loads and video consumption gives way to “all the dials turned to 11.” Someone spending an hour watching Reels has less time for Netflix. Sharp initially doubts familiar AI-video displacement arguments, then accepts this cost-pressure mechanism as more persuasive.
11. State power shapes AI access and military infrastructure
Responding to a listener who understood him as demanding Anthropic serve the government, Thompson clarifies his argument. If AI becomes civilization-altering, governments will assert authority over its deployment: “They have the guns.” He is describing power and inevitability. The separate normative debate over surveillance or weapons targeting does not resolve whether a private supplier can dictate terms to a state possessing overwhelming coercive power.
Thompson distinguishes commercial Starlink from military Starshield. “As I understand it,” the US military controls Starshield’s use; he acknowledges uncertainty about the precise boundaries of Musk’s authority. On Ukraine, he notes that the country was using commercial Starlink rather than Starshield and raises the possibility that Washington asked Musk to enforce the service’s terms.
Both hosts stress how little they know about the reported Ukraine slowdown. Thompson guesses Musk was in close alignment with the Trump administration; Sharp guesses the US government asked him to do it. These remain hypotheses. Their shared point is that Ukraine has less power than Washington or the infrastructure supplier it depends on.
Thompson’s industrial argument is that consumer demand supplies military scale. Defense primes serving one government customer tend toward expensive, low-volume systems; drones and missiles increasingly require volume. His analogy is car factories repurposed for wartime production. His proposed route to domestic drone capacity is categorical: “Ban all foreign drones and completely legalize drone delivery across the US.” Sharp embraces the commercial-volume argument.
12. Persuasive filmmaking can carry a disputed premise
Thompson calls You Can See Everything “the Zootopia of documentaries”: excellent craft carrying a premise he finds untenable. He rejects a sympathetic reading of Elizabeth Holmes as someone manipulated by men and interprets her relationships differently. His concrete objection is the theater’s alarm at Holmes granting Billy Evans power of attorney, which Thompson regards as normal when someone enters prison and a partner cares for the children.
Sharp’s defense rests on the film as commentary about media and truth, rather than as an Elizabeth Holmes documentary. He sees multiple layers of construction and interpretation and thinks it should win best picture. Thompson grants the filmmaking’s persuasive power while remaining frustrated by what that persuasion accomplishes.
13. The Brewers’ title rewards an outlook built around the season
Thompson had made peace with the Brewers’ small-market limitations and learned to enjoy summer evenings without expecting a World Series. Basketball with Giannis felt different: a real chance creates relentless pressure. The Brewers’ victory initially produced surprisingly little feeling because he had “cauterized the postseason in my heart.” He then becomes emotional explaining the unexpected reward of enjoying both the season and its championship ending.
The connection back to AI is uncertainty and control. Thompson objects to people depressing themselves about an unknowable future while missing the life they can live now. “Be nice to people around you. Learn new things. Explore new things.” His sports lesson is that outcomes lie beyond a fan’s control; his broader advice is, “Figure out the future when it gets there.”
Full transcript
Hello and welcome back to another episode of Sharp Tech presented to you by Stripe. I'm Andrew Sharp. Joining me in studio today, a world champion, Ben Thompson. Ben, how you doing?
I’m doing fantastic. I did not want to talk about the Brewers. You brought it up last week. I was like, “No, we are not mentioning it.” But now—
We can mention it.
I’m free. We can lean into it. And you know what? Sharp Tech colors, Brewers colors. Who knew?
Here we are. I was going to be very disappointed in you if you did not wear a jersey to our taping today. You did me one better: you wore a jersey and a Brewers hat.
Well, the hat is the one piece of, you know—
Oh, it’s a World Series hat.
A World Series hat. So I had to get—
Doing it right here.
I have not had a chance to get the World Series champion merch, so I’ll address that on Friday.
I trust that you will. I can’t imagine how much money you’re going to spend on World Series merchandise on the heels of that win.
I do have takes.
I know you do.
Let’s save them for the end. Actually, I’d like to request the entire last segment. I have 2 takes: I have a Brewers take.
Okay.
And I have—I went to see a movie.
Oh boy, Ben went to the movies. That’s a good teaser for the end of the episode. You did mention you could soliloquize at length about the Brewers, so I look forward to that when we get to the end. For now, we’ll begin with Anthropic’s S-1, which went public at 9:00 a.m. Monday morning this week. We’re going to try to do the impossible, which is make a fun podcast out of a 400-page document drafted by lawyers and accountants. I know that’s fun for you, though.
I can’t decide if it’s fun or not, to be honest. It was a real challenge. It’s been busy these last few weeks. You asked a couple of weeks ago—or did you? I think John asked—is it going to drop this year? Once I read it, I was like, of course it’s going to drop this year. They’re actually in a really favorable window to drop it, as far as their business goes. But dropping it the week of the World Series, when I have games to watch—
People actually don’t understand how much you’ve been juggling with the Brewers’ playoff run over the last month or so. You’ll watch every playoff game and then write from 1:00 a.m. to—
6:00. We don’t talk about that because I don’t want people judging my output. But yes, I am relieved about the Brewers in multiple dimensions.
We don’t have to do this anymore.
That’s right.
I’m really thrilled that we have some actual numbers to discuss with any of these frontier labs. Dario opened the hood. We get to look at the engine a little bit. There were 93 pages of that document dedicated to risk factors. I’m actually less interested in risks and that whole conversation that’s surrounded Anthropic for the last couple of years.
Well, there is one. Did you see this sort of satire tweet going around about Roko’s basilisk?
No, I didn’t.
All right. So, Roko’s basilisk. There’s a Twitterer, I think, and the idea is that you should be careful about what you’re writing because the AI is going to consume it.
And just in case it reads it and knows what you think or how you act—
Okay.
—you should be positive about the AI.
Oh.
So it’s kind of like—there’s some philosopher who has the same—
Keeping a list of naysayers.
Well, there’s a thing about why you should be a Christian or believe in God. I can’t remember what it’s called, but it’s the same concept: you might as well, because in case you’re wrong, the expected—
Downside.
That’s right. It’s very high. Someone put up a screenshot of text from an S-1 document saying, “Wow, Anthropic’s risk factors are off the hook,” and it had a whole thing that was like, “Our models may be aware of what you did.” So, if you don’t invest in our IPO, the model may take that personally or whatever. Number one, it was very funny. Number two, it was believable.
That’s what I was going to say.
There were a couple of different parody risk factor sections floating around on Twitter this week, and literally anything somebody threw in there, I believed. Who the hell knows what they put in the risk factor section? But again, that’s sort of been the conversation with Anthropic for like 2 years now. I think it’s pretty interesting that we have real numbers. I have some specific questions for you. Topline question, though: I saw someone tweet that Anthropic’s IPO will be for the AI trade what Coinbase was for crypto—essentially, the top of the market, and it all goes sideways from here. Based on what you saw in the S-1, do you buy or sell that take?
I don’t know. I mean, there’s no question—no, there is a question. I think it’s inevitable that we are in some sort of bubble right now, and it’s going to pop. It’s going to be painful, and then we’ll come back. That’s just the inevitable nature of these things. I think there’s an extent to which AI is different in lots of areas. There are lots of differences we’ve talked about between the dot-com era and this era, in terms of the mechanics of things piling in too quickly and a mismatch in timing between when it’s absorbed or not. But that dynamic—where the dot-com era had lots of good ideas that just didn’t make sense for a while, or didn’t get penetration for a while—I think that’s almost inevitably going to be the case.
Okay. Now, when does that happen?
I don’t know. There have been a lot of issues with the neoclouds.
Neoclouds, yeah.
Right. Can they get financing? I think we were pretty on the ball with this. We’ve been talking about the concern about the credit mismatch, not just whether we’re going to run out of money. This is when we had the discussion about railroads. We talked about Google. I know we’re going to get into some of the financing questions later on, but I think it’s likely there’s going to be pain.
Mm-hmm.
But—
That is the case, and people will experience pain. My experience reading your coverage of the S-1 this week, though, was that Anthropic actually has a rock-solid business.
Yes, you’re totally right. The core business looks amazing. The subscription is definitely profitable. It’s less profitable than the API, but—
The subscription is a great product. People are like, “They’ve already paid for it, so they’re going to use it.” They’re going to get used to your product and your workflow, and they’re going to advocate for it in their companies.
Claude people, you know, Claude loyalists.
And once you’re paying for it, it’s like, “Of course I’m going to use that.” Now, I think it’s worth it to pay for both, but who’s actually going to do that? By the way, someone figured out the whole point system Claude applies to you, reputationally—
Wow, I did not know Claude was bestowing social credit.
Apparently, if you sign up—well, I think it’s kind of for some of the distillation issues or questions, or whatever.
Ah.
So if you sign up and do the Max plan immediately, you start in the basement.
Interesting. Okay, I have to find—
Counterintuitive. Perfect.
Well, it kind of makes sense if you think about it. I think—yeah, I’m pretty sure I have a one. Maybe I started out in the basement.
So the business looks amazing. On one hand, that’s a reason not to show that off.
Yeah.
Companies don’t always want to show how good their business is, for competitive reasons.
Mm-hmm.
But that’s actually not the concern in this case. It’s great. It’s a great talking point. Everyone realized, “Holy cow, this is an amazing business.”
Right?
The problem is the R&D spend is much larger than the cost to serve, and it’s increasing more quickly than the cost to serve, even in the first half when their usage is exploding.
Right. In that half year, the cost of their R&D spend for training and all that sort of thing went up more than the cost to serve.
And the issue is: why did it go up? Because they want to make a better model. Why do they need to make a better model? What’s the whole point of pacing the frontier?
Because there is competition.
Because there is competition. And what that means is the largest portion of their cost is not, in fact, inference. The largest portion of their cost is this training cost.
And the speed with which they need to spend money on that is out of their control.
Right.
It is a matter of competition.
It’s an ongoing cost. It’s not a one-time cost.
Huge cost. You crack the code and then get it under control.
That’s right. And so that is the concern here with this company: they have a cost function that is kind of out of their control. All indications are that the bigger the models get, the better they get, the more it—
Costs. More expensive.
And this is where we circle back to “We need to pace the frontier for safety.” It’s like, we need to pace the frontier to make sure our company stays in business. That’s definitely the case.
Yes. Well, as an example, their ideology just lines up perfectly with their business needs.
Well, as far as losses are concerned, another number floated before this was released: The Wall Street Journal reported that Anthropic had a loss of $80 billion for the first half of 2026, based on the S-1. Can you unpack that figure and why it’s not as ominous as it sounds?
It’s not ominous at all—though it depends how you look at it. What happened was, when Amazon made the investment of a lifetime, along with—who’s the SBF?
Oh, SBF. Yeah.
He lost his; Amazon kept theirs. So when Anthropic was counted out completely, and everyone thought, “OpenAI has won. This is just second place coming along. Nothing’s going to happen,” Amazon came in with a brilliant investment. They got multiple things out of it. Number one, they said, “You’re going to use Trainium.”
Yeah.
Did Anthropic want to use Trainium?
Absolutely not.
Probably not. That turned out very well for Anthropic, by the way. I didn’t write about it this week, but I think I’ve written about it in the past. They talked about it in the S-1: Their compute is built from the ground up, and their whole model is built from the ground up to be heterogeneous and chip-agnostic.
Yeah. They’ve built a certain level of flexibility into their overall cost structure.
Because they had to. Their very first models were on Nvidia. Then they had this whole Trainium thing. Once you’ve already sort of supported two, then okay, now we’re on TPUs, and now they’re going to be on AMD. They have a deal with AMD. They’ve built a layer of abstraction that’s a very good place to be, and I think it’s benefited them very well. They’ve similarly always been on all the clouds.
Like, they’re always on Amazon, then Google, and now they’re on Microsoft. I think that’s very underrated in why they took over the enterprise under OpenAI’s nose. OpenAI was stuck on Microsoft.
So what advantage did that give Anthropic with enterprises?
If you have all your data on AWS, you’d rather use the service on AWS. There’s another thing—sorry, it’s an S-1, like I was saying.
400-plus pages. A lot of threads.
About 50% of their revenue is coming from the Big Three via those services: AWS, Azure, and Google. It confirms they’re doing kind of sketchy accounting, depending on how you define it. If you pay Amazon $100, Anthropic is recording that $100 as revenue.
Anthropic revenue. Okay.
In reality, about a third of that—people were saying about 37%—goes to the platform provider, whether that’s AWS, Azure, or Google. They’re putting that 30% into sales and marketing.
I was wondering what the sales and marketing costs are. So that’s part of it. Okay.
A little bit of BS accounting, in my opinion, for 2 reasons. Number one, I think it’s fair to argue that they shouldn’t recognize that 30% at all. They should only recognize the portion that’s going to them, so their revenues would be lower.
Yeah. And when we say 15% or 85% margin, that’s totally excluding this portion.
If their revenues were commensurately lower, the margins would be compressed because the costs are still the same.
Or number two, I’m not sure how that’s not a cost of goods sold.
Because it is a portion of the revenue. It’s not like you’re paying a fixed fee. As you do more revenue, the portion you have to pay the platform scales as well. From my perspective, I think that should be in cost of revenue.
This is what I mean, by the way: You live for this. You love accounting.
I absolutely love it. I’ve got a Brewers hat on right now, but I’ve always got my accountant’s hat on. I need a pocket protector. I needed to get a custom Brewers jersey with a pocket protector. That’d be great.
Amazon’s $80 billion—and that—
We’re getting there. You’re weaving. You’re weaving.
I am weaving. That’s right. So those numbers we talked about for margin aren’t quite correct, because there’s a significant amount that should be accounted for. It’s in the financial statements, but—
I think this is more of a cost of goods sold. The margins are lower.
Still great, to be clear. We’re not complaining about the margins. That part of the business looks fantastic. So, you go back to Amazon way back when: They had to be on Trainium.
They were obviously going to be part of AWS, but also Amazon, I believe, basically gave Anthropic a bunch of money.
Yeah.
And this money was actually given in the form of a loan.
Anthropic could take that money and buy Trainium compute. It’s a nice little circular financing arrangement, which people complain about. But from Amazon’s perspective—and we can get into the role of the hyperscalers here—it makes a lot of sense. They’re basically double-dipping in the advantage.
Well, not just that: Anthropic using Trainium makes Trainium better. They’re going to push the limits of what this architecture can do and develop more tooling for it, all of which makes Trainium a better product to offer to anyone and everyone.
So this is a massive win for Amazon, which they got because Anthropic had nowhere else to go. They needed lots of money and lots of compute. Microsoft was with OpenAI. Google was doing its own thing. Amazon stepped in.
So, as part of this loan, Anthropic could pay back the money, or Amazon could take shares.
Right?
And that was Amazon’s call.
Now, of course, Amazon is going to take the shares.
Yeah. But every time Anthropic raises money, the value of those shares is repriced, and they have a liability to Amazon. I believe they’ll give the shares at the IPO. They raised money multiple times, I think, in the first half of this year, and every time they had to reprice this, it was booked as a loss in their accounting.
So that $80 billion is fake. It’s not money out the door.
The beautiful thing about accounting is everything has to balance, right? One thing goes up, another thing has to go down. That money shows up on Amazon’s income statement as the increased value of its investments.
Yes. And it’s a reflection of the astronomical growth Anthropic has experienced. This is actually going to be a big problem with all big tech earnings going forward, because all the hyperscalers have various investments in these companies.
Their earnings are going to go up and down based on the value of these investments. One of the challenges Google had with its earnings this year—we’re not going to talk about Google’s earnings; we’ll probably get to it next week—is that the value of its SpaceX holdings went down. That impacts its earnings. You’re going to have to look at operating income, not overall income. That’s going to be the case with Amazon in the opposite direction: Its overall income is going to be way higher because of its Anthropic investment, which doesn’t actually have anything to do with understanding the underlying business.
Okay. So we can all look forward to that. We’ll see.
Super interesting and also very annoying.
Yeah. Some of these, like Google, have had their investment in Anthropic reflected in massive earnings over the last couple of years as well, right?
For sure. Yeah. This has been sort of a roller coaster. Once these companies become public, that’s actually going to be kind of fun. And, by the way, there are a bunch of these outstanding. Anthropic has this sort of loan to—I can’t remember what these are called, but there are more with Amazon. Amazon hasn’t finished converting the first one. They have 2 more.
Is this from 2 different sets of—Anthropic is paying for compute as much as anything else?
Of course, yeah. But this is the whole thing: Who’s actually paying whom?
The investors? The existing investors?
Well, yeah. They’re basically materializing this money out of thin air, which means every other investor in Anthropic is heavily diluted by these. That’s what goes out. This is the whole thing about it: Anthropic is just super aggressive about this, but this is the case with all the hyperscalers. It’s the case with everything as we’re moving into equity for funding this stuff. The people who are paying are existing shareholders. You’re just devaluing what you own. And this is built into Anthropic. Number one, they have all these outstanding share issues. They have them to Broadcom, they have them to Google, and they have 2.5 more to Amazon, where it’s just built-in dilution. They also have it built in that if the value of our stock goes up by, like, 50%…
Then we have the right to issue 2% of outstanding shares for this benefit. Sure, it’s like you’re just—
The fun continues.
Yeah. I mean, I guess they’re very straightforward about it.
No, but dilution is baked into the equation, because there was a tweet from someone named Vincent Jansen on Twitter. He wrote this week, “First, hyperscalers used free cash flow. Next, hyperscalers used debt. Now hyperscalers are turning to share dilution. Is there a final step, or is this the last?” Anthropic is very much part of the same story there.
Yeah. Well, again, we talked about this on Sharp Tech this summer. I think we were very ahead of the curve. This is the last, but it’s also, in some respects, the best.
Mm-hmm.
Because the reason companies want debt first is that they retain the upside.
Yeah. Once you’re giving away—or once you’re having to sell shares—you’re basically saying, “Give us money because you believe, and there’s going to be—”
And you’re giving away the upside to new people, to new people buying those shares—
Which I think, in this space, is—
—the healthiest, all things considered. So you could go glass-half-empty and say, “Well, now the burden is being shifted to retail investors of the future to fund these bets.”
Right.
You know, I am sympathetic to the private investors who got all the upside to date in Anthropic. If you were a retail investor, you could have had money in Anthropic and sort of seen this coming. It’s not like it was unknown that they had these deals. But now the public can buy in when we’re worth $1.5 trillion or $2 trillion. Where are we going to go from here?
Their proposition is pretty straightforward: Look, you can buy in. It’s super expensive. We’re going to dilute you endlessly, but we’re going to control the world. And how much is that?
We’re going for a ride. Absolutely. My counter to somebody who made that argument to me would be that it’s preferable to the scenario where Anthropic just remains private in perpetuity and does it that way. I think it’s healthier to have public-market disclosures, number one, and also public-market upside in the midst of all this.
One thing that’s interesting is there are a couple of timing things here. It was a year ago that Dario was mad at, or mocking, OpenAI—not mocking, but saying they were reckless for buying all this compute.
And literally 6 weeks later, he looks like a dummy because they don’t have enough compute.
And all of it’s more expensive now.
It’s more expensive, and so now they’re like, “Actually, we’re going to get a ton of compute.” I suspect—I can’t say for sure—but I suspect that made the first half of this year look even better financially.
Because their costs are compute. They didn’t have enough compute.
Ah.
But their costs were lower even as their—well, because this is a similar idea: Who’s actually paying? The people who were paying in the first half of this year were the people for whom Anthropic was down every few hours because they didn’t have enough capacity. All those people did their part for Anthropic’s IPO by making their margins look—
—like they were winding through their frustrations with Claude.
That’s right. Well, the other thing that’s interesting is these numbers in the S-1 are as of June 30.
That’s what I was going to ask about. What are you still wondering about as we move forward?
All the reporting—and is that in whatever data we can see, like Ramp’s index of spend or whatever? Did it sort of flatline—not flatline, but did the rate dramatically decrease in the second half of this year? Astra comes out, OpenAI is really, really pushing, and then, on the flip side, Opus 5.5 came out. So did it reaccelerate?
This is all, I think—it expires. What’s it called? It doesn’t dehydrate; it goes stale. I think it’s November 12, and the reporting is they’re going to go public in mid-November, right? Will we get fresh numbers to celebrate?
We have to get fresh numbers as of November 12.
Okay. Are they going to try to get out before then, or after that? Is there going to be an October surprise? It’s interesting, though, because the narrative around Anthropic—people are very bullish on this company in any context—but it’s become a bit more nuanced over the last several months because of the progress OpenAI has made in the enterprise, and because of random Ramp charts that show growth flatlining, at least in relative terms. Given the growth over the last couple of years, they’re just not going to be able to replicate that growth going forward, because it’s impossible. It’s the law of large numbers. So let’s see. We’ll see what it looks like.
But the counter to this is that was, I think, the accepted wisdom a month ago.
Mm-hmm.
Opus 5.5 is so good.
Like the—
Fair enough. I mean, maybe it’s going to say, “Yeah, it flatlined, but then it took off again.” The only problem, if that’s the case, is that it reinforces that we have to spend, because all that matters is how good our model is. If all that matters is how good your model is, you have this built-in, accelerating cost center in your business that you just have to be sprinting on the revenue side to outpace.
You had this question for Shyam Sankar in your Stratechery interview this week: Is it inevitable that the labs have to come for all software companies because of the commoditizing nature of their models? I think about Anthropic’s costs, and obviously competing on the frontier is going to continue to be incredibly expensive. That makes me curious about what other lines of business they try to come for to realize all their upside in the years to come. Maybe that allays some of the concerns about continuing to compete on the frontier.
Yeah. I mean, you have all these lines of battle in tech. One is obviously OpenAI versus Anthropic, right? That’s the headline.
That’s been the big blockbuster fight for 3 years now.
But there’s the frontier labs versus everyone else. I think the two companies on the forefront of the “everyone else” are Microsoft and Palantir, both of which are making very similar propositions to their customers and making the argument publicly. Satya Nadella and Alex Karp have been doing this: You need to be in a place where the model is completely replaceable. Everything you do needs to be around it—
—and they’re model-agnostic.
They’ll trot out every argument they can come up with for why this needs to be the case. There’s a lot of, “Anthropic will steal your business, too,” as part of the rationale. A big part of it is that if you give them your data about how your business works—
Like—
—it’s only a matter of time.
It’s like, what do you have?
But another one is that you should be able to sub them out. Why do you want to sub them out? Because you want whoever gives you a cheaper deal, whichever model is better. So there’s a real fight and tension going on here.
The argument I’ve made—I asked this to Shyam, or told him, “I apologize for calling you Shams.” I’m hurt.
You did drop a “Shams” at one point.
Oh, did I? Oh, no. I told him at the end, “I kept wanting to say Shams.” Unfortunately, he’s a classic tech guy; he didn’t know what I was talking about.
Completely lost on that one.
Was Shams Charania a newsbreaker in the NBA?
I think at one point in the interview you dropped a Shams bomb. I don’t know—is Shyam Sankar known for bombs?
Yes. No. He’s known for writing these emails to the whole company or a whole team, just taking them to task. He’s like an enforcer. They’re called “Sham bombs.”
Okay.
But no, it wasn’t really. I did get a long email from him, but it was nice.
Yeah, he clarified there weren’t enough expletives in his email to qualify.
So, yes, Shyam Sankar. I’ve been raising this for ages. I think in the long run, in the fullness of time, they need to go up the stack, integrate into the customer, and own the customer experience. You can even see this with things like Claude Code versus Codex. Working with Claude Code and working with Codex are different; they’re organized differently.
The way I think about it is that Codex is like people who file their emails into folders. It’s very project-based, and you have threads within projects. If you’re that sort of person, very organized, it makes more sense.
As a former lawyer, I think of Codex as the associate and Claude as the partner.
Yeah. Claude is better at handling lots of different things. You’re constantly changing topics. But if you want to be organized on your own, it’s kind of maddening. Needless to say, that’s why I use Claude a lot.
Claude’s kind of a dick. The personality still drives me crazy. But for me, the way it ends up manifesting is that I work on this—I talk about GeckoBot. The actual building of it, where it’s clear I’m building this piece, this piece, this piece, I’ve done mostly in Codex.
And the actual model powering it is Claude, because it just handles different things coming at different times.
Now, it is kind of interesting in its own right. I'm almost choosing from a UI perspective as opposed to a “what the model does” perspective. But the more you have that choice from a UI perspective, UI is essential, because that's the interface. It's the interface to humans. It's literally the name: user interface.
But you have people who are used to, say, Claude Code, and they don't want to go to something else. Right now, you can sub stuff in and out to a certain extent. You can call different models, so that can be worked around. But the further up they go in the stack—or say you do a Codex work project, whatever it's called, I can't keep track—you get your little virtual machine in the cloud, so you have your own VM. It can write stuff down. It's like, wow, it's so helpful: they gave me this virtual machine in the cloud.
Yeah, you're now stuck in their cloud. Yeah, right. Or in the Claude Design thing, you know, what they did to Figma kind of did them dirty. But the more you're using that—using an application—the less likely you are to switch.
Locked in. That's right. In a world where all you need to do is change an API, that's the world to go back to at the beginning, where you're on this treadmill and you don't have very much pricing power. That's right.
Sure. Well, and it does. I mean, you could talk yourself into limitless upside if you talk yourself into the possibilities of Anthropic just eating SaaS over the next couple of years and doing all sorts of different things in the software area to expand its revenue pie. And maybe they could do that. I mean, revenue pie—it's lock-in.
By the way, lock-in is—
—an important part of making money. Like, what's your moat?
Right. Well, right now it's very difficult to falsify any hyperbole about what Anthropic or OpenAI could accomplish in that space. For now, we have a couple of emails on Anthropic.
1. Anthropic's Obsession with OpenAI
Alexander says, “Andrew and Ben, I watched Dario and Daniela's interview this week with Walter Isaacson, and woof. They start off the interview by praising the inherent honesty and goodness of Anthropic, then spend the majority of the 42-minute discussion criticizing OpenAI for having ads in its product, building more sycophantic models, partnering with the government as ‘an opportunistic move’ rather than a ‘principled stand,’ and being generally dishonest. The level of obsession, hatred, and disdain they have for OpenAI is alarming.
“It's like someone badmouthing their ex during their wedding vows. Someone should remind Dario and Daniela that when you point one finger at someone else, there are 3 fingers pointing back at you.” Not sure how that works. I hadn't heard that phrase before, but—
Are you serious?
I mean, I'm pointing a finger at you. There's no—oh, there we go.
People will have to watch the video for that one. But sure enough, there were 3 fingers pointing back at me.
I am baffled how there's a question here about the Gen X–millennial thing. Is this a generational divide? How have you not heard this before?
It could be. We're only divided by about 6 years, but maybe that's enough of a chasm to create that disconnect. In any event, what do you think of Alexander's sentiments there? It was funny you included this email because I was literally thinking the exact—
This is the hazard of recording podcasts. Sometimes you're just thinking out loud on air for everybody. But there you go. Learn something new every day. All right, so you were thinking the exact same thing that the—
It's kind of pathetic in a way.
Okay. The extent to which they continually define themselves as the opposite of OpenAI. It's kind of funny because they're like, “Oh, about alignment and training our model to be great.” If they continually make decisions like, “Oh, we're not going to do what OpenAI does,” who's actually setting the agenda here? If OpenAI is too online, Anthropic is—
Yes, anchored on what they're doing, right? Sure.
And to what extent might OpenAI actually do something that makes sense, that is the right thing to do, and Anthropic is like, “Well, I guess that's the wrong thing to do”? I do think it is a very astute point. It speaks to the overall discomfort I have with their entire—
—view of the world. Okay, which is like, this stuff is too powerful, so someone has to teach it how to be good, and the only ones that can be trusted to do that are us.
And you always ask, “Why you?”
And Alexander pointed out what their answer is: because we're not Sam Altman. It's like, dude, if your definition of being good is not being Sam Altman—
—you—
Well, and maybe I'm too immersed in this world at this point because, to me—
Well, you're still defined by Sam Altman. It's like people who define themselves as being against Donald Trump. You have let Donald Trump—
—define the entire frame in which you're operating. But it's almost just ambient noise to me when I hear them go off on one of these tangents about how untrustworthy OpenAI is or how untrustworthy Sam Altman is. I mean, there's an element of this where it's more charming than some of their other crazy digressions. Their obsession and their rivalry with OpenAI isn't really that big a deal one way or the other, as far as I'm concerned. I can't get too worked up about it. Except these are the people setting themselves up as the moral arbiters—
—for everything.
So I don't believe that. So I've never taken them seriously.
No, I believe they're trying to do that. The answer isn't to—
Just like, I don't think it's the right approach for them to teach Claude morals.
I'm not here to teach Anthropic morals. What I'm here to do is make sure that there's no small collection of people that get to decide it for everyone else. Yeah, that's the bigger—
And to the extent that the mudslinging at Sam Altman is intended to delegitimize him in the eyes of the mainstream and ultimately redound to Anthropic's monopoly power over how this technology is deployed.
That sucks. It’s also, so far, been—
To what extent has Anthropic actually been the best thing for Sam? Like, oh my God.
Right, like suddenly—
And they’ve done wonders for Mark Zuckerberg as well. Everyone who’s worried about Anthropic is like, “Sam, he’s not that bad. I didn’t figure it out, so—”
“We can work with this guy.” Well, when you go back to OpenAI, when Dario was there, the original beefs with Altman, at least in some cases, were—
The dumbest stuff OpenAI did was from Dario Amodei and his whole team: the whole “GPT-2 is too dangerous to release” thing.
Right? Like this, this—
Well, and part of their falling out involved a clause that would have required OpenAI to merge with another company that was closer to AGI than they were. Sam wouldn’t include that clause, or Microsoft wouldn’t agree to including it in the investment. Dario was very upset, like, of course Microsoft wouldn’t agree to it, and of course Sam wouldn’t make that a dealbreaker for the—
Seriously, this does go back, though, to the original sin, which is this whole, “We’re going to be a nonprofit.” Actually, companies should operate like normal companies.
Traditional incentives.
And this is the whole problem with the S-1 and Anthropic’s insane governance structures. We haven’t talked about that either. There are, like, seven share classes. There’s this magic share that the board has, or the benefit corporation has. What are these random things about who gets voting rights and who doesn’t? It’s nuts.
Like, actually, we should just operate like normal companies. That would be—
If you’re buying in, that’s what you’re buying. Yeah. Well, we’ll see.
Profit motives.
Good, healthy—
Good for society. Healthy, yes.
You know, I was writing about China for Sharp Tech this week, and they have all these political incentives that kind of distort things, obviously, famously. America, as crazy as America can become in the eyes of the world, is a place where the rest of the world can trust that there are basic profit and market incentives guiding the decisions that are made. That’s incredibly important. It is its own rule of law.
And it gives you—why is the rule of law important? You can make decisions. You can make investments—
Within a predictable framework.
That is what the traditional corporation, the profit-maximizing corporation, gives you. Yes, there are obviously issues. They’re well documented. But every system has issues. The benefit is that you have rational actors. You can actually engage and understand someone else. This seven-share weird governance, where we’re not actually governing for profit, is worse, and—
I mean, you know what you’re signing up for, to be fair. You’re signing up for, like, “We’re going to just dilute you away, and we’re not focused on making a profit.”
And maybe they’ll pull it off because Opus 5.5—
Because at the end of it, that’s what we always come back to: the core product is really, really good. All right. Ravi says, “I was just listening to Sharp Tech at 3x speed while on the treadmill. I usually listen to podcasts at 3x, but for Ben’s updates or interviews where there’s also a transcript, I listen and read, which is even better.”
Yeah. See, Ravi, you should have filled us in.
There we have it. I don’t watch video versions because I don’t have YouTube Premium and can only go to 2x.
Ravi—
Does he need YouTube Premium? That’s a hilarious reason not to watch video podcasts, because it only goes to 2x.
I mean, look, you can live your life the way you want to. I’d rather you pay your subscription money to Stratechery versus YouTube if there’s only one.
He’s making good choices as far as I’m concerned.
Yeah. You know what? I have nothing to say. Continue with the email.
2. AI Bubble Questions
Great interview with Craig Moffett. He asks, “Where does Ben land on his key question about AI capex?” This is Moffett’s key question: Is end demand from consumers and enterprise where it needs to be, and will that be sufficient in the near term, in 2027 or 2028? The direct quote from Craig Moffett: “All I care about,” he said, “is whether the revenues coming into the system—which means coming into Anthropic, coming into ChatGPT—are greater than the cost of building the whole value chain. Ignore all the revenues going on between the layers of the value chain. That’s just noise. All that really matters is enough money coming into the system to support the system.” Where do you stand on that question? I thought it was a pretty good formulation of the core issue from Craig Moffett.
Right. It’s easy to focus on how much Nvidia is taking, how much TSMC is taking, and what’s up with the neoclouds. Yeah.
No, it’s a very good observation, and I don’t know—we don’t know what the timing is right now. As impressive as Anthropic’s revenues are, they don’t come anywhere close to all the capex that’s being put out. But the capex is also for 2, 3, 4, or 5 years down the road. If Anthropic keeps accelerating the way they are, then—accelerating based on what time frame? From the beginning of the year? Later on? This all matters. Their revenue growth is the other point.
This goes back to people getting really worked up about stock-based compensation, because that is also dilution. They’re like, “You should be evaluating this in your stock picks.” The reason people don’t care about stock-based compensation is they only care if your revenue isn’t growing. If yours is growing, no one cares. It’s fine. Once yours is growing, they’re like, “Yeah, well, and also, have you seen their stock-based compensation?” Right? Like this.
And so all this stuff I talk about with Anthropic—the dilution, the weird governance—no one cares if it’s up and to the right. Yeah, and that matters broadly for the whole ecosystem. The thing to come back to, though, is that Amazon deal at the end of the day, and the question that we got about going from free cash flows to debt—
—to equity. The entities with the most capacity in all three areas: who has the most cash flow? Who has the greatest capacity to issue debt because of their large cash flow and diversified businesses, which gives them collateral—basically, not collateral, but other sources of income?
Yeah.
And who has the greatest capacity to issue stock? The hyperscalers.
Google, AWS—or Google, Amazon, and Microsoft. Yeah, there is an extent to—
If we cross the gap, they’re like the girders of the bridge that get us across. Okay.
And it’s going to be very interesting to see how they play their hand over this little bit, like to what extent does it end up—
So Sundar Pichai is holding up the bridge? Well, I mean, are we going to end up in a world where Anthropic is over here doing all their crazy stuff, and what actually happens is AWS and Google basically own them to get them over the canyon? Okay.
And, of course, we could reach a limit. The limit where everything goes sideways is once the hyperscalers can’t do it anymore.
But there they are. What would that mean? They have the cash flow of last resort. They have the debt-raising capabilities of last resort. I mean, even the neoclouds, all this sort of stuff that’s out there—the hyperscalers’ cost of capital is so much lower than—
A neocloud trying to raise money. NVIDIA’s out there—
Cloud. Yeah.
Or, yeah, the neocloud. Yeah. Why is NVIDIA backing up the neoclouds? This is one of the concepts. This is when I really wrote about the credit crunch, or the credit questions, and the railroads. Was NVIDIA putting backstops under all these neoclouds—
—to help—
NVIDIA sees this happening too, right? Because—
The worst-case scenario is the world runs out of money before the revenues catch up.
And that’s what happened with the railroads and the massive crash. Basically, the world just ran out of money. The rails weren’t useless, right? The rails are still used today, but there was a massive financial crash. The way we don’t get there is we just have to bridge the gap to where the revenues are enough to pay this all back.
And the challenge for NVIDIA is, if we get there, it’s probably because of the hyperscalers.
Okay. And that means the hyperscalers have gained tremendous power over the entire ecosystem, because you get what you pay for, right? In that world, those are the ones who are going to build competitive chips to NVIDIA and put pressure on NVIDIA’s prices, all these sorts of things.
When we say “tremendous power,” is that because Anthropic and others will have to continue trading equity in exchange for compute going forward?
Right. Yeah, and the terms Amazon got with Anthropic were when Anthropic needed it the most—
Because they were trying to get started, and no one believed in them. They needed a ton of compute, and so Amazon got an unbelievable deal—
Right?
You can fast-forward. The deals today maybe aren’t going to be quite as good, but if Anthropic is like—
“We can’t operate.”
“We can’t operate.” No one—our equity, like, if their stock price starts plummeting. Yeah, the neoclouds, they have to turn and ask, “Where can we get money—”
“To be competitive?” Amazon’s like, “I’ve got some money for you.”
Remember that deal we made in 2023, 2024? How about an even worse one?
Right? Like—
I can’t wait to see what happens over the next couple of years here.
Well, and I think Moffett’s core question is, is the money for any of this here? He doesn’t doubt the technology, but it’s how much people—how much enterprises are willing to pay, and whether that can sustain any of this over the next couple of years, which I think is something of an open question.
I mean, yes. If you want to really zoom out, I think one of the biggest broader questions—and I really enjoyed the email with Shyam this week—
—and one point he made up front is from Palantir. He’s talking his book, of course, but—
Of course, but I mean Palantir—
—he admitted in there, because I kind of pushed him about it, that cult recognizes cult. Part of the reason they’re worried about Anthropic is, “We’re the right religion; they’re the wrong religion.” But what they see with companies focused on using AI to improve the bottom line—
—struggle, because how do you improve the bottom line? Replacing humans.
Yeah.
How do you improve the top line? Multiplying humans.
Right? There’s a difference between multiplying and replacing. When you’re multiplying humans, you’re creating entirely new things to do, new workflows, new opportunities. The challenge for this whole trade, in general, is that usually, number one, it takes longer—
—and number two, it usually comes from new companies—
—that come in and disrupt the incumbents: the AI-first company taking on the old-school SaaS company. The SaaS company is trying to shove in AI and use as much of it as they can, but they’re fundamentally not constructed properly to do it. In the fullness of time, those companies go away. The AI-native companies take their place, and that’s how AI takes over the world. I think there’s a very decent chance that’s how it happens.
Mhm.
The problem is, it takes a long time.
Yeah.
And timing is the issue.
All the money being spent on capex is being spent now.
Yeah.
And that money needs to circle back, because when you go from paying with free cash flow to paying with debt and paying with equity, the idea is that it circles back around to your cash flows paying it back, right? So the question is, how long is that going to take? It’s all about timing. I 100% believe in AI. I think that’s pretty straightforward. It’s incredible.
I’m in the middle ground. I don’t think AI is going to replace everyone. It’s going to be good, but—
But it’s just unbelievable. I was just showing someone this week.
It’s so fun when people suddenly unlock. They’re like, “Holy cow.” Once you see what it can do, you come up with 25 new things it can do.
I could not be a firmer believer in this. And yet, I said this years ago: the greater the upside of something, the longer it almost takes.
Mhm.
Go back to the internet. The feed wasn’t invented until 2006.
Yeah.
The core unit of the internet, the core monetization unit of the internet—the feed with personalized advertising. We got search ads along the way, but even then, we’re talking 1999 or 2000. There’s a lag between the new technology and understanding how to monetize it, the new businesses it enables, the user interface, and the business model around it.
Also, while we’re here, if this whole interview is them saying OpenAI is evil because they include advertising, okay. It’s a personal shot at me. That’s all I’ve got to say.
Oh boy.
But I think it’s a point I’ve made before, and it’s worth repeating. Again and again, I feel there’s an alignment between optimism about AI and optimism about advertising business models.
Mhm.
If you believe in advertising, why do I think Meta advertising is good advertising? Because it shows you stuff you didn’t know you wanted. It’s not just a win-win-win. People create businesses that didn’t exist, and I get stuff I didn’t know I wanted.
It grows the pie.
It grows the pie. But it’s manufacturing human desire.
Mhm.
There’s a certain—what’s the word I’m looking for? An ascetic perspective, like not eating anything, or being very low-key, where I was appalled by consumerism and all these sorts of bad things. Fine, whatever. But consumerism, at the end of the day, is what drives the economy.
In a world of AI, the whole mindset that AI is going to completely replace humans rests on the assumption that the entire economy is utilitarian: everything is done because it needs to be done, and AI is just going to do it better. I don’t think that’s the economy. A huge portion of it is status-seeking. It’s buying stuff you don’t need. A huge portion of it is trying to get laid. It’s spending an unbelievable amount of money on World Series tickets, which I may not have done. That’s what it means to be human, and it’s inextricably tied to advertising.
Yeah.
It’s not an accident that one of their go-to points about why OpenAI is evil is that they’re including ads.
It speaks to how Anthropic is deeply an anti-human company. They don’t get it. They don’t get humans.
Socialist company there.
There’s this mindset in Silicon Valley that’s anti-ad.
I think it’s rooted in this idea that people are just getting tricked. It’s very utilitarian: “I can figure it out.” It’s a communist, top-down view of your needs. Not necessarily—
Right? And it can be controlled, right? We just haven’t had the tools yet. But now AI can do it, right? So we can actually understand all the inputs. This is a lack of faith in the market, a lack of faith in humans. All this talk about no one having a job is fundamentally a lack of faith in humans.
Sure.
It’s like we can’t come up with things that might be worth paying for that AI won’t do better.
I bet we can.
I bet we can. I can think of some right now.
That’s right.
We have.
So it’s funny that that’s one of their talking points about why OpenAI is evil.
Yeah.
And it’s one of my talking points about how you don’t understand humanity. The audacity to think we’re going to create this consciousness for a machine—it’s insane. These people are nuts.
I mean, Alexander’s a better man than I am. I don’t think I could slog through 42 minutes of the Amodei and Walter Isaacson interview. But it’s not surprising to me that they’re slinging mud at OpenAI over ads. Whatever. I don’t really take it seriously, so it doesn’t bother me.
Bloomberg reported this week that Meta shares fell after Mark Zuckerberg told investors the company would bet even more aggressively on artificial intelligence. During its quarterly earnings report Wednesday, the social media giant boosted its capital spending plans last Wednesday. This was a week ago, which is a pertinent point I’m going to get back to, and it announced a new class of nonvoting stock that could ultimately be used to help pay for acquisitions and other expenses.
The Bloomberg story references a plan to raise up to $95 billion in waves of stock offerings as Meta continues its cash-raising spree to fund AI. Meta shares are down about 10% in the last week. How much worse would the reaction be without the success of Muse?
Well, just one thing to note is the shares should be down there. People are being diluted, right?
One of the interesting things about the dilution is that companies like Meta, Google, and Microsoft can probably get away with it. Who else do you feel really comfortable investing in in the present-day economy?
That’s my whole point from before. That’s why they’re the last resort.
And if you’re lending money, who else do you want to lend it to? Meta’s got—
We’ve got foreign governments sinking lots and lots of money into these companies.
Yeah. The reason I mentioned the timing is that I was annoyed. I thought, “Okay, it’s fine. I can do Meta earnings even though the Brewers game is tonight. It’s a company I know well; I can knock that out.” Then Anthropic drops its S-1, and I’m like, “I’ve got to cover the S-1 before I get to the earnings.”
Yeah. I’m so glad I did the S-1 before the earnings, because I think my entire take on the earnings was based on what I learned in the S-1.
Mhm.
Which is this point that actually all the money is going to training.
Yeah.
That gives a very clarifying perspective on Muse. We’ve said on this podcast—I think we had a Muse run for a couple of weeks in a row, at least two in a row—what an unbelievable opportunity is here: the aggregator of aggregators, the interface, the actual connection to the user, the layer between everything else. I talk about Stripe being the platform of platforms. This is like the aggregator of aggregators, on the opposite end of the sandwich. It’s a very valuable place to be. But to be there, the model has to be good.
Okay.
And you have to control the model. It’s the key component of it.
To have a good model, to control your own model—
—you’ve got to train your ass off.
—you’ve got to spend a lot of money. And so this is the question for investors who are questioning their spend: if you believe in Muse, you should be happy they’re spending money.
Mhm.
If you don’t want them to spend money, then you need to appreciate that your take is actually that they should not be doing Muse.
Yeah.
They're inseparable.
Mhm.
And I believe in Muse. I believe in this layer. The overall issue with Muse right now is that the model isn’t good enough. That’s the number one issue, even above—
I don’t really experience that, interestingly enough.
Well, I mean, the other issue is all this interface layer between companies and pushing back on agents and all that sort of thing. No—
3. A Brief 1Password History
They got the 1Password integration finally. That’s a game changer for me.
By the way, 1Password—little tip to the builders out there. You can use 1Password as your secret store if you’re using agents to build. There’s a whole issue of secrets like API keys, usernames, and passwords. How do you pass them to the agents? You don’t want them to put them into the chat; they’ll go up to the servers, and it’s not great.
You can do things in a terminal with a secrets file and all that sort of stuff, but it’s super hard to manage. 1Password has this amazing layer where you create a vault for the agent. You can do it on a per-app basis, and the agents can access the vault to get the keys they need.
And then, from a user perspective, I’d rather type usernames and passwords into 1Password than—
—doing convoluted terminal commands and pasting stuff. Absolutely. Yeah, it’s great. 1Password is in the E-ZPass, Shop Pay tier of tech products, as far as I’m concerned. I completely agree. A 10 out of 10.
No, they’re one of those sort of random—I don’t want to say random—but a company that, if you’d predicted who was going to be massively well-placed for AI, you probably wouldn’t have thought about. But it turns out they shifted to being an enterprise product. They’re another classic example in the Dropbox genre. They started out as a consumer product, and a lot of Mac people—I can see Jim over here. I think he’s grimacing because there’s a whole area of people who were so mad—
People.
They’re so—no, no. They’re so mad at 1Password. It used to be very custom-built, with beautiful native apps and all that sort of thing.
I missed that era.
No, and so they became more web-appy. I’m not sure that’s completely true, but more web-appy. What’s the word? What’s the framework everyone complains about? Electron, or whatever. That’s Slack. They basically made the app worse. Yeah, the app got worse, but it became more extensible, easier to work on, and cross-platform. Their whole business model shifted from consumer to enterprise, and they raised a bunch of money. So all the old 1Password people—
They’re still mad about it.
Right? But that’s what happens. If you sell a product, if you sell software, you sell to businesses; they’ll pay. Consumers require advertising. We talked about this whole thing before. Well, because you also have Apple coming in, making its password management more and more capable and actually having an app around it. So it’s hard to be a consumer—
—software company.
—if either your ads or your operating system is coming for you. Tough place to be. They moved to enterprise. Everyone gets mad at them. Brilliant move. Number one, it was a brilliant move in general because that’s a better market. The password area in general is so difficult to manage. We’ve been using it at Stratechery for ages, in part because we have to share passwords for a lot of different stuff. Super useful in that regard.
You can enforce rules. The—but number three—
Turns out, for agents, it’s amazing. It’s a company that was sort of out of—
—nowhere, and it’s a really big value unlock in terms of how I use Muse. So, in terms of the investment—
That’s a perfect example, right? One of the big problems with Muse is: how do you get it your password? Suddenly, if you share it there, there you are. It’s all there. But all this to say, Muse—there’s no question whether you think the model’s good enough or not—
4. Muse as the Ultimate Founder Bet
Meta has to control that model, and they can’t be outsourcing that to Anthropic or OpenAI, which, by the way, are going to want to own the interface as well.
You could just cut them off, right? They have to—they have to control the model underlying Muse, which means, as we saw from Anthropic’s S-1, they have to spend a ton of money on training, even more than I fully internalized.
And this did give me some sympathy for their enterprise thing. It’s still optimistic. They’re setting up an enterprise company because they’re spending so much money on training and need to sell it somewhere.
Find money somewhere. Sure. The reason it made me feel a little better is that the concern about doing enterprise is that you’re taking compute away from all your other products. It turns out serving this stuff doesn’t take that much compute, relatively speaking.
Yeah.
So—
This is what I mean: it’s great to have Anthropic open the hood.
It’s true. That’s why it’s true.
Makes it a little clearer. And the other good thing about Anthropic coming along is that this is their business.
And I think, like I said, whoever at the SEC made them break out this compute—
—the whole document was worth it for that table, because the other companies, like Meta and Google—
—this has just sort of grown inside their balance sheets, so they’ve never had to break it out.
Totally.
And so now you can take what you learn from Anthropic’s S-1—that must apply to Meta’s S-1. We’ve spent 4 years speculating about the cost breakdowns here. I will say, you wrote the most compelling argument for investors opposed to this spending: not that Muse doesn’t need it, but that Meta doesn’t actually need Muse.
Yeah. But—
I’m not mad at Meta for going this direction. Mark Zuckerberg is a founder who was always going to go this direction. This seems a lot more fruitful than some of the other detours of the last 7 or 8 years.
The investors want Meta to be Google or Amazon. Yeah.
Companies that are not run by founders, that—
You know, does Amazon—they—
—are just printing and optimizing, and that’s—
They are focusing on what they’re good at, what makes sense. I think Amazon is in a great position. This isn’t to say they’re doing the bad thing. I think Amazon is in a great position as far as being the inference provider, with all the infrastructure, all the logistics, all the things we’ve talked about, and just focusing on that—not spending billions and billions of dollars trying to be at the leading edge of models.
Last summer, they kind of wound down their effort. They claim they’re still doing it somewhere else, but I don’t believe them. Why would they? They are there; they’ll serve Anthropic, they’ll serve OpenAI, and they’re built to be a lower-margin business anyway. So they’ll have a place for this.
Smart for where they are. Well, the Google one’s interesting. Again, we’ll save Google for next week. Google for next—
—but they are prudent, rational decisions that you would expect from a professional manager.
Yeah.
Meta is not run by a professional manager. It’s run by Mark Zuckerberg, who founded the company.
Yeah.
So, like, what do you expect?
Totally.
And also, I’m all in. Let’s go. Let’s do it.
Well, I just can’t help but observe that their ads business does look pretty bulletproof.
Well, that’s the other thing: of all the digital things in the world, Reels feels the safest. It’s also eating everything else, from the other side—
They’re eating Netflix. We talked about that.
Interesting. So here’s a take for you.
Okay.
Netflix is a casualty of AI.
How so? I’ve heard others offer that thesis, and I never quite buy it.
What is the other thesis? Because I think mine’s better. But—
I’ve heard people argue that people are going to start creating their own videos. No, I—sorry, I have a 3D, 5D, 6D. Yeah. So, you ready?
So what’s the—so Meta is spending all this money on AI.
Okay. They’re obviously under investor pressure. They need as much money as possible. You go back 8 or 9 years ago, they used to talk on earnings calls about not wanting to overdo the number of ads in the feed—
Right? Because we want to make sure the user experience is good. They had Facebook Video, which they reduced, because they were like, “It’s not good for users to get consumed by video.” They had a period where their time on service went down because they were deprioritizing video, and it sounded like—
—and it’s almost like that was their monopoly phase. They could just do what’s best for people, do what’s best for society because it’s the right thing to do, the better user experience. We don’t have to worry about time. What are they doing, right?
They need money. They’re shoving ads in there. They’re ramping up the engagement. They have all the dials turned to 11: making you watch more Reels, making you never look away, see more ads—
—which means your eyeballs—your wife is spending her—
Reels time.
An hour.
And what is she not doing?
Watching Netflix.
So Netflix is actually a casualty of AI because AI is pushing Meta to ramp up engagement.
5. Elon's Role in the Ukraine War
That’s interesting. I do buy that, honestly. As the cost pressure intensifies for some of these companies, Netflix is going to have a harder time. I like it. All right, well, that’s enough Meta. I’m sure we’ll come back to Meta in the Google episode. We do a lot of Meta on the show.
Which, by the way, is in part because Mark Zuckerberg’s in charge. They’re more interesting, in some respects, than some of these other companies.
They are. I mean, certainly over the last couple of years, they’ve been super interesting. And I love Muse. Utkar says, “I’ve been listening to Sharp Tech since the beginning. I was very surprised to hear the opinions of you and Ben saying that Anthropic should essentially be forced to offer its products to the US government, even if it doesn’t agree with how they might be used.”
Okay, sorry. I knew you were going to jump in as soon as I finished reading that sentence. What do you got for Utkar?
For clarity, the point of the Anthropic versus DoD article—
—is that, in the fullness of time, if AI is as powerful as they say it is, they will ultimately have to deal with the entity that has a monopoly on violence.
Yeah.
And that’s a reality people don’t want to think about or be honest about, but need to be. The point is, they can defy the US government. You can make the case that if the US government doesn’t want it, they shouldn’t buy it.
The reality is—
If they’re talking about civilization-altering technology—
The government is going to have its say.
Yeah.
And they have the guns.
So it’s not that I think this is good. This wasn’t a normative judgment. I was being descriptive.
Yeah. You drop “normative” a lot. I’ve never really used “normative.” I’m not quite sure if I’m using it right.
I’m not saying it’s right or wrong. I’m saying it is what it is.
It was—and it was very controversial, obviously.
It’s just one of those things that, even at the time, was so obvious to me.
But it was so annoying at the time that we had so much pushback on this, that you’re advocating for it.
So I’m not—I’m not advocating for it.
Yeah.
I’m stating what is—
Going to be a fact of life.
Right. And the public-private tension is not going away in a number of spheres.
Utkar says, “I personally agreed with Anthropic that AI shouldn’t be used for mass surveillance or weapons targeting, so I sided with Anthropic, as I thought they were doing the right thing.”
Again, Utkar, I’m not disputing your opinion on this matter, but your judgment is operating at a level that is not the level I was discussing. It wasn’t whether this was right or wrong. That’s an important debate to have: Should AI be used for this? The debate I was having was at a higher order, which is you telling the US, “You have access to this tool, and we’re going to dictate to you what you can or cannot do.” Number one, that’s fairly insane on its face. Number two, if you fast-forward, and the models increase like you say they will and become as powerful as you say they will, it’s somewhat intolerable now. It will be completely intolerable. Again, politics is violence without guns.
Wait, is that the way it goes? Violence by no other means?
Something like that. I can’t remember.
Well, yeah. Looking back at that, there were so many insane opinions at the time about the administration, the American AI system, and the ecosystem. This was part of the problem: people were like, “You’re disagreeing with you.”
I wasn’t arguing that AI should be used for surveillance. I was saying that argument is immaterial.
Right.
Anyhow, maybe we’ll wait for another email in three months that says I said something different.
Well, I think your piece aged well. I wrote something that also aged well: “The End of the World as We Know It.” Fast-forward to last week: Utkar says the Wall Street Journal reported that Musk ordered the slowdown that blinded Ukraine. I don’t know for sure if he acted completely unilaterally, but it’s now clear to me that a private company should not be able to unilaterally decide the foreign policy of the United States—and, in this case, basically the entire world, since it has no viable competitor. I’m wondering, based on this, have you changed your opinion on Starlink? I can’t imagine anyone in Europe is happy about this, even if they can’t really do anything about it in the short term. We have not talked about this. Sounds like you came around to say I was right all along. Sorry, cars. Now I feel bad. Jeez, unbelievable. I’m a terrible person. What’s your reaction to what may or may not have happened with Starlink and Starshield, Ukraine, and the Trump administration?
So I don’t know what happened with this. It’s profoundly opaque, and so dangerous to podcast about because we just don’t know. As I understand it, number one, something that’s right about Starlink is that there are 2 services.
There’s Starlink and there’s Starshield, right?
Starlink is a public—private, commercial, whatever word you want to use—open-to-the-public commercial offering. Starshield is a military offering.
Mm-hmm.
Starshield, as I understand it, is under the US military’s total control in terms of how it’s used.
Right.
Musk, or anyone else, cannot dictate how that’s used—or will not, I’m not sure what the lines are, but at least according to Musk. I would imagine that’s the case. I think that’s a good division. It makes sense, hearkening back to the best ideas of dual sourcing. We get mad at China for dual sourcing; I think there’s a little bit of jealousy there. Dual sourcing is obviously good because you get volume from the consumer market, so you get a better product for the government and important capabilities for the government.
Right. We do a bit of that here as well.
Of course we do. The entire Arsenal of Democracy—the whole World War II machine—was repurposed car factories turning out planes.
Civil-military fusion, sure.
The whole problem we have right now, in losing our industrial base, is that we don’t have that capability. What do we switch to making all this military stuff? Our military contractors serve only the US government, so the US government is stuck with these insane entities that exist basically to lobby it. And then you tend toward very low-volume, very high-cost weapon systems, because you’re serving only one customer. It’s like the subscription business: You want to maximize your revenue per user because you don’t have as many users. If you’re selling only to the government, you want to maximize your revenue per item because you have only one customer. If you’re serving the entire market, you make it up in volume. The future of war, in my estimation, is clearly things like drones, missiles, and unmanned systems—a volume play. None of our defense primes are built around volume plays.
Yeah. Do we need to fix the regulatory regime to enable the flowering of a domestic drone industry for the sake of national security?
Yes.
If we’re playing sixth-dimensional chess, maybe we do.
No, actually, from a military perspective, it’s nuts that we don’t have drone delivery.
Yeah. How?
Ban all foreign drones and completely legalize drone delivery across the US. You’re going to get your drone capabilities. That is actually the route to do it.
Yeah. Anduril is out there as the new poster child. In many respects, they’re just building the idea that if you want to form a defense tech company, you have to go through them because they got into the Pentagon first. And, by the way, if you talk to people in the Pentagon, they’re like, “Yeah, the Anthropic stuff needs to actually work, or the Anduril stuff needs to actually work at some point.” That’s sort of the pushback there. God bless Anduril and Palantir. They were there when it wasn’t cool. But what you need is volume. Great take by you. I like that. There you have it: drone delivery tomorrow, US military dominance next week.
But the Ukraine affair—did it change your opinion in any way? This is where you don’t know who decided what.
Yeah.
As I understand it, the terms of service for Starlink define where it can be used. I know this for sure on a couple of levels. Number one, you have to buy different levels of plans. I have a Starlink on my roof for my house as backup. That’s a relatively low rate because it’s fixed.
Yep.
I also have a mobile Starlink that I take to Canada. A couple of weeks ago, I went to a baseball game and set it up so I could watch the Brewers game at the same time. I pay more for that because it’s mobile; it can be used everywhere. These are just basic contractual terms-of-service things. They have to know where the Starlink is because it aims itself at the satellites.
So they know where it is. Another example is that there’s no Starlink in Taiwan. I was talking to a buddy who was up on top of a mountain or something, and the internet wasn’t great. I said, “Starlink’s awesome.” Nope. It doesn’t work in Taiwan.
Has Musk ever said why that is?
As I understand it, the Taiwanese government is the issue here.
Okay.
But, yeah, obviously it’s a hairy situation in general: China.
I have some questions.
I think there’s probably—well, this is actually a great example. Whose fault is it that there’s no Starlink in Taiwan? It’s in Musk’s interest for the Taiwanese government to be against Starlink, so he doesn’t make China mad. Right?
You get to Ukraine. The U.S. doesn’t want, as I understand it, Ukraine making attacks deep into Russian territory—
On oil refineries.
Right. Right.
Right, pushing up the price of oil.
Mm-hmm.
Which Ukraine was doing using Starlink. So is there a, “Hey, Elon, can you enforce the terms of service,” which Ukraine was violating?
They’re not on Starshield. They’re on, like—and that’s balanced by the fact that Starlink has saved Ukraine’s rear end. Again, I don’t want to get into the whole thing.
This is why it’s so murky. You can see who’s pushing on whom. At the end of the day, they’re using a consumer service that a U.S. company has no loyalty to.
There’s a lot of the world right now that’s just things that were—
The U.S. has no loyalty to—
A U.S. company has loyalty to the U.S. government.
Yeah.
They don’t answer to the Ukrainian government. There’s a little bit of presumptuousness sort of embedded here. But, again, we’re not a political podcast.
Yeah. To me, broadly speaking, it’s a good object lesson in why these public-private dynamics are potentially dangerous, in terms of what happened with Ukraine. My guess is that Musk was in pretty close alignment with the Trump administration and the U.S. government in terms of what buttons were being pushed.
My guess is the U.S. government asked him to do it.
Sure. And he’s falling on the grenade to a certain extent. But if it were a Democratic administration—here’s the point to your overall point: We’re talking about power.
Yeah.
This is the whole thing. Who has power?
Mm-hmm.
Once you grok—no pun intended—that we’re talking about power dynamics, you get what I was going for with the Anthropic versus DoD article. It wasn’t about legalistic niceties. It wasn’t about mass surveillance. It was about who, at the end of the day, has power.
6. Ben Sees You Can See Everything
Absolutely. And if Anthropic’s not going to jump when the government says jump, then the government is going to investigate other options and ensure its suppliers aren’t relying on Anthropic. It is going to continue to be an issue, though, because particularly as everything becomes increasingly polarized, the questions get awfully thorny. But in any event, we’ve gone too long. There were 2 segments that you demanded we do on the podcast, so—
Wait, this is unbelievable. We’re an hour and a half in. Do you— I wanted—
No, no, no. First of all, we had a little interruption about 10 minutes in, so we have an extra 5 to 7 minutes here. The text message that you sent me 30 minutes before recording was you looked at the rundown and said, “Why no You Can See Everything segment?” with multiple question marks. So—
What else did I say?
I think you said you were terrified of my takes or something to that effect.
You’re trying to muzzle me.
Trying to muzzle you. I don’t want to muzzle you. All right, what do you have for me? You went to see You Can See Everything because of the World Series. You saw it at 10:00 a.m., I believe last Saturday.
More than that, because of the World Series, I promised my daughter, who was very jealous of me and my son going—I might have gotten a few other games in this run. I said, “If they make the World Series, I’ll fly you up to go to a World Series game.” I didn’t know what I was getting into.
But yet, you were a dad of your word.
That’s right. And so she came up.
Mm-hmm.
And she was insistent that I see it. She wanted to see it again. It turns out the local theater had added an extra run at around 10 in the morning, so we could get to it.
Then make our way. Did a full tailgate. The weather was beautiful. Parking lots were open.
For a second, I thought you tailgated the movie. Great. I’m glad you got out of the movie in time for a tailgate.
You Can See Everything is the Zootopia of documentaries. People may not have the context: you weren’t really a fan of Zootopia.
Zootopia is a very entertaining, very well-done movie.
Okay.
That is insane.
The premise of the movie—
Great.
This is an old Sharp Tech segment, but the premise is—you’re like, “Oh, have you seen K-Pop Demon Hunters yet?” I—
Which you have to do, because I ask you everything. But this was the deal on the last show, and I almost went to see K-Pop Demon Hunters. I almost sat down with Netflix, and I was like, “Can’t do this. At some point I’ll do it, at some point I’ll do it.” At one point I did say the only cartoon movie that I’ve enjoyed was Zootopia. Then Ben proceeded to break down why it’s just a completely ridiculous portrait of society.
It’s not ridiculous. It’s ruined it for me.
The premise of Zootopia is that animals whose nature and instinct, because they’re carnivores, is to eat other animals, are bad—
They’re benign.
No, it was saying they’re bad. The bad actors were the ones who wanted to eat other animals.
Okay. It was saying that their fundamental, intrinsic nature made them evil—
And that they should just magically change their digestive system, everything about them, to live happily ever after—
With coexistence with other animals in the animal city. Yeah, man.
Extrapolate the meaning of this broadly. It’s an insane view of the world that we’re somehow mainlining to our kids.
Okay.
Terrible movie. Very well done. Terrible movie.
And you know what? I’m not arguing with you on that point. When I watch it through that lens, it is very frustrating.
You Can See Everything: this blank-slatism from Zootopia somehow leaves you sympathetic to Elizabeth Holmes, this poor, well-meaning little girl being taken advantage of—
By the men in her life. Absolutely. Elizabeth Holmes is a psychopath. She is a liar and a whistleblower abuser. She’s been in these relationships before with Sunny Balwani, which, again, was framed by some as him taking advantage of her. When you see what happened with this Billy Evans guy—whatever it is—it sounds a lot like this relationship. It made me rethink what I knew about Theranos: she consistently finds men she can manipulate to basically do the cover for her, to play-act at being a scientist—
And live in this mystical reality that has no connection to the real world. You’re not buying into Theranos 2.0?
Hearing her talk about the movie is like, holy cow, she has no idea what she’s talking about. This lady is crazy.
So, what was the big one where the whole theater gasped?
Ooh, I don’t know. I think there were a couple.
Well, there were a couple, but one of them was that she signed over power of attorney.
Okay, yes. And I was like, “Oh my God.” That was a point about what a manipulative guy he is. Turns out, what do you do if you’re going to prison and your spouse is taking care of your kids?
You probably should sign power of attorney.
It is totally standard, totally normal, because guess what? You’re in prison.
You’re going to blow out the studio’s mics. Second episode ever.
So it’s unbelievably well done. It’s well done because if you don’t know what it’s about going in, he convinces you 100%.
Yeah.
So the craft, like Zootopia, is amazing, and the premise is nuts.
So it is enervating for you to sit there, having known more about the backstory, and see it portrayed that way.
Yes.
Okay.
Absolutely insane.
So did you enjoy yourself across the 3 hours at the Cineplex? I mean—
I like that I have no blood on my hands, because this was your daughter forcing you to go to the movie. I was very confident that you were not going to see it for, like, 3 years after our conversation. What did you think on a scale of 1 to 10? Just the experience of it.
The first part was—it definitely got better once you went to prison.
Okay.
Like—
It turns into a different movie at that point.
Yes, it does.
I love the pregnant pauses.
I can’t decide. I’m not really a movie guy, so I felt a little guilty subjecting this guy to 3 hours in a theater. Well, I think, as a sort of postmodern commentary on media and truth, that’s the only way you can embrace the movie as a success as a documentary, because of the holes that you just highlighted.
It’s almost insulting. And there are all sorts of crazy things, like a well-known conservative figure who was assassinated.
That’s one of the other lines in the movie where the whole theater is like, “Gasp,” for sure.
So I’m in Madison, Wisconsin, which is very progressive, to say the least, and the reaction in the theater was, “Ah, the gotcha moment that means Billy Evans is a bad guy.”
Okay.
The whole thing at the time: a guy got shot in the neck, and—
Oh wow.
She’s happy about it.
That week, right?
Wow.
So I thought that was fascinating. It was fascinating not as me putting myself in your shoes or making a commentary on that point, but because that was arguably the peak of the movie—
—that was itself a meta-commentary on top of meta-commentary, because that specific event and the reactions to it—
—was the ultimate societal divide.
Yeah.
One side is like—
—very specific to that time period. I didn’t even—
—get shot. One side’s like, “He was a racist,” and the other’s like, “What does that have to do with him getting shot?” It was just a fundamental divide in how you see the world. That specific episode and the way it’s played out as shorthand since then.
Yeah.
The fact that that was the climax of the movie—
—is, to your point, building on why you like the movie, an even higher meta level on top of all the levels of meta that were going on in the movie.
There were about 8 different meta layers at play throughout that movie. I think it should win Best Picture, but again, it’s not an Elizabeth Holmes documentary. Definitely not how it should be understood by the mainstream.
Zootopia.
I’m just glad we got you out to the theater, man. Once in a blue moon, we’ll see you at the Cineplex.
7. Emotional Thoughts on the Brewers World Series Win
All right, now we’ve come far enough. Let’s talk World Series. Can I begin as a friend of yours? The fact that the Brewers won this series is endlessly delightful to me as your friend, because I’ve watched you sink into this baseball obsession over the last several years. You’ll watch basically every Brewers game throughout 162 games, and I’ll joke with you, like, “Wow, you’re really a Brewers diehard.” Over the last couple of years, I swear to God, we’ve had about 5 different conversations where you’ve said, “I’m in it for the journey. We’re never actually going to win a World Series. That’s not what it’s about for me. It’s about summer evenings watching my team every night. The rhythm of it is what I’ve come to love. I’ve accepted that we’re never going to compete with the big markets.” And now here we are, a World Series champion sitting across from me. I just love to see it, man.
You just took my whole segment.
Oh, really?
I’m sorry.
No, it’s true. There’s the whole thing about, “How can you be romantic about baseball?” Baseball is a metaphor for life.
And yeah, that was the thing.
You’re getting emotional.
I’m getting emotional right now.
No, I’d given up because it did seem inevitable. Something about basketball is different. Basketball is very binary.
One player can make a difference.
That’s right. And so you have Giannis. The pressure is insane. We have to win it, but you 100% know you have a chance to win it.
With the Brewers, you get to the postseason, and you have to get homers from the most expensive guys. We’re never going to have those guys because we’re a small market.
The sport is not set up for a team like the Brewers.
That’s right. And so you had to enjoy it. Otherwise, what’s the problem with the late Giannis years? It’s all or nothing. There’s a 10% chance that this could work.
Right, and it almost makes it less enjoyable. The whole path of a basketball team: the early years are the best because there are no expectations. It’s almost less fun when they’re winning, because it’s a grind. Once you get over the hump, it’s, “Can we get back?” It’s miserable. The last few years of a dynasty are kind of miserable to cheer for. It was different with the Brewers and baseball, in that it’s like a life lesson. How do you avoid nihilism about the whole thing?
Mhm.
What it really is about is embracing the journey. Enjoy the journey. Enjoy the summer nights. It’s okay; there’s a little tournament at the end of the year. It doesn’t really matter.
Yeah. And the great thing about this is it was a weird reaction when they won. When the Bucks won, I was in tears. When Khris Middleton hit that shot, I was in tears as a Giannis guy. And here I was, almost like I didn’t feel anything, in part because I’d cauterized the postseason in my heart to such an extent that even winning, I could barely feel anything. But it was great. It was a really pleasant thing that happened to me. I’m getting emotional again. The beautiful thing is I got the best of both worlds. I enjoyed this whole season without thinking we were going to win the whole time, and then they won. What a better way to live.
You weren’t anchored on, “Can they do it this year?” because you knew the answer. I love that you’re actually crying.
Let’s break this video out. No, but rather than spending July, August, and September obsessing over that question—
—you had answered that question. The answer was no.
That’s right. I’d answered that question: it doesn’t matter what happens at the end. It’s not going to impact me right now.
Mhm.
We’ll get to the end when we get to the end. And you think about it with AI: everyone’s worried about the future that’s unknown.
Yeah.
We’re having debates about it. How about you enjoy life right now? How about you live your life instead of always thinking about the future or the past? Live in the present. Be nice to people around you. Learn new things. Explore new things. Figure out the future when it gets there. The reason is, number one, you don’t have control. This is the thing about sports: the worst thing about being emotionally invested in sports is you have no control.
Yeah.
All you have control over is yourself.
I have no control over what the Wizards do.
You have control over those season tickets.
But that is why this isn’t just great for sports. It’s great for life. Live in the moment.
Yeah. I enjoyed every game of the Brewers this season. It was an awesome season. People online, on Twitter, always throw it back in your face—Cubs fans in particular—like, “You know, you could just lose in the postseason.” It’s like, yeah, we probably will. And that’s fine. Then we get here, and we also won. I get to enjoy a double dip. I had a great season. I had great playoffs.
The Brewers are world champions.
You’re going to a great parade on Friday afternoon.
The Brewers have all kinds of fantastic merch for you to spend money on.
That’s local sports.
It’s really great.