The $800 Billion Catch Behind Anthropic's IPO

Anthropic

Across the shows: 62 speakers in other episodes, 48 bullish and 44 bearish calls, past 30 days

• Paul Kedrosky · SK VenturesBearish

  Anthropic's operating loss is narrower than expected, but the business is nowhere near profitable and will worsen.

  The first-half operating loss of about $693M was narrower than Kedrosky expected (he had pencilled in roughly $1.1B), but he reads it as a best-face number prepared for the IPO and not a durable trend. He says the company is nowhere near profitable and will likely get worse for reasons he expects to change. Profitability would need compute intensity per unit of token output to keep falling sharply, since otherwise compute costs would consume the business. The filing's roughly 50% figure for compute-intensity decline is the metric he thinks has to keep improving.

  “full credit to them that I think this is nowhere near profitable and it's probably going to get worse in future”at 7:00

• Paul Kedrosky · SK VenturesBearish

  Anthropic's heavy stock-based compensation will dilute public shareholders and force large buybacks.

  Kedrosky points to stock-based compensation as a material add-back in the filing and says Anthropic is very generous with it, a pattern likely to continue after the IPO. Dilution from that compensation would require substantial share buybacks to offset. That would make it a real claim on future cash flow, with the buybacks competing with funding for compute commitments.

  “Anthropic is very... generous in terms of stock based compensation, and that will probably continue post IPO.”at 7:48

• Ed Elson · Prof G MediaBearish

  Anthropic's compute commitments above $800B are off the income statement, so reported losses understate future spending.

  Elson adds up the Broadcom and Google compute agreements in the liquidity section and says Anthropic is about to spend more than $800B on compute on a timeline it does not disclose. None of it appears in the income statement. He concludes that the income statement tells you little about the business's coming cash needs, since the costs have not yet hit it. He does not argue this is deliberate; he sees it as a timing gap.

  “they're about to spend more than $800,000,000,000 on compute. And none of that is included in the income statement.”at 9:47

• Paul Kedrosky · SK VenturesBearish

  Anthropic's reliance on rival clouds and noncancelable compute deals will force costly direct builds and shift cash flows.

  Anthropic distributes its models through Microsoft, Google and Amazon, which take a fixed rake on sales and could cut it off, so Kedrosky compares the setup to a streaming service with a fixed cut and rival competitors. The S-1 says capex and upfront lease payments will rise, which he reads as a move to build its own infrastructure. That shift creates large, noncancelable capital commitments, and he expects maintenance capex on top. He says the historical income statement is therefore a poor guide to the business's future cash needs.

  “Because in particular, it's not just the size of the commitments, they're all noncancelable.”at 10:53

• Paul Kedrosky · SK VenturesBearish

  Anthropic's revenue growth is decelerating and must reaccelerate sharply in Q4 to be credible.

  Using leaked Q3 numbers alongside the filing, Kedrosky estimates revenue growth has slowed to about 50%. Management argues Q4 will show a spike, which would imply growth of roughly 64%. He says a company listing in this position needs to reaccelerate growth to be credible. He treats the implied reacceleration as a test the company has to pass, and he is skeptical that it will.

  “The more troubling piece is that there is a revenue deceleration underway... They're now looking at something like 50% revenue growth, but arguing that in Q4, they'll see a spike.”at 18:05

• Paul Kedrosky · SK VenturesBearish

  Anthropic's IPO valuation will likely land near $1–1.5 trillion, not the reported $2 trillion.

  Asked whether the IPO will price at the reported $2 trillion, Kedrosky says that if it goes ahead it will land near $1.05 trillion. He adds that the IPO may not happen at all, because he expects the US government to stay shut for the rest of the year. The transcript is garbled on the exact number, so the figure is uncertain. He gives it as a point estimate of about $1.05 trillion, possibly $1.5 trillion, with a probability of zero if the government stays shut.

  “I think if it goes, it goes to $1.05 or so”at 18:55

Space Exploration Technologies

Across the shows: 34 speakers in other episodes, 27 bullish and 15 bearish calls, past 30 days

• Paul Kedrosky · SK VenturesBearish

  SpaceX's cancellable compute rental revenue is being presented as AI revenue, but it is not.

  Kedrosky says SpaceX buys and resells compute, roughly $3B a year, on 90-day notice, so the contract is cancellable. He thinks SpaceX counts this as growing AI revenue even though it is a rental of compute to someone else. He reads the arrangement as SpaceX wanting cash flow from it, and says the revenue is far less durable than the label suggests.

  “they, in turn, consolidate it as growing AI revenues when it's really not, obviously.”at 11:30

CoreWeave

Across the shows: 52 speakers in other episodes, 12 bullish and 47 bearish calls, past 30 days

• Robert Armstrong · Financial TimesBearish

  CoreWeave goes to zero if AI data-center financing dries up and the bubble deflates.

  Armstrong argues that if the AI data-center bubble deflates and investors stop financing that capacity, the hyperscalers can survive by returning to free cash flow. Smaller neoclouds, which depend on outside funding, would be hit hardest. Musk's warning that debt markets have cut off companies without independent AI-data-center revenue supports that financing squeeze. Armstrong says CoreWeave in that scenario goes to zero.

  “CoreWeave's not going up. CoreWeave's going to zero under that.”at 33:42

Broadcom

Across the shows: 5 speakers in other episodes, 4 bullish and 2 bearish calls, past 30 days

• Robert Armstrong · Financial TimesBearish

  Broadcom would lose massive value if AI data-center spending deflates, as part of the chip complex.

  Armstrong groups Broadcom with Nvidia as part of the semiconductor complex that has become important to the AI build-out. If the AI data-center bubble deflates and hyperscalers stop financing capacity, he expects those companies to lose a massive amount of value. He calls that a shock, though a smaller one than if it hit Microsoft. The view depends on the deflation scenario, which he does not forecast.

  “Those things would lose a massive amount of value.”at 34:20

NVIDIA

Across the shows: 41 speakers in other episodes, 34 bullish and 14 bearish calls, past 30 days

• Robert Armstrong · Financial TimesBearish

  Nvidia would lose massive value if the AI data-center bubble deflates, though still the world's most valuable company.

  In the scenario where the AI data-center bubble deflates and hyperscalers stop financing capacity, Armstrong says the semiconductor complex, including Nvidia and Broadcom, would lose a huge amount of value. He calls that a shock, though less of one than if it hit Microsoft. He notes Nvidia is currently the most valuable company in the world, so it has the most to lose. The argument rests on the scenario, not a forecast that it will happen.

  “Those things would lose a massive amount of value.”at 34:20

AI infrastructure & data centers

• Robert Armstrong · Financial Times

  The AI build-out will go through restructuring and consolidation of neoclouds, as in past railroad booms.

  Armstrong compares the AI build-out to a railroad boom in which everyone financed a line. He expects it to go through restructuring, reconstruction and consolidation, and says that will happen one way or another. Musk positions himself as a consolidator who will buy distressed rivals, but Armstrong says Musk's financing record is the main reason he might be able to do it. He believes the consolidation will happen whatever Musk does.

  “This first round of, like, a railroad boom... This is gonna go through a round of restructuring and reconstruction and consolidation.”at 30:49

• Robert Armstrong · Financial Times

  Hyperscalers' core businesses are very profitable and could regain free cash flow if AI data-center spending deflates.

  Armstrong argues that Alphabet, Amazon, Meta and Microsoft entered the AI spending cycle with some of the most profitable businesses ever. If the AI data-center bubble deflates, investors could reward them for going back to free cash flow, buybacks and dividends instead of data-center spending, and those stocks might rise. He expects the big stock market pillars to stay standing even as others suffer.

  “Four or five of the hyperscalers that whatever happens to AI, they went into the situation having the most profitable businesses in the history of the world.”at 32:56

US Treasuries & yields

• Robert Armstrong · Financial Times

  The bond selloff was overdone; 2–3% real yields and a 5% ten-year look attractive.

  Armstrong notes that the ten-year and 30-year yields rose while stocks wobbled, then fell about 19 basis points in a day as investors rushed into Treasuries. He thinks the earlier bearish mood about the bond market was overdone. Inflation-adjusted yields of 2–3% are, in his view, pretty good. A 5% nominal ten-year yield with some real yield gives a cushion if inflation rises a little more than rates, though it will not make you rich.

  “the bad feelings about the bond market may have been overdone.”at 24:06

US stock market outlook

• Robert Armstrong · Financial Times

  Markets look like the end of a cycle; the euphoria is broken, though he isn't calling for a crash.

  Armstrong says the Nasdaq and S&P 500 have moved sideways since May. Beneath the index, AI-linked industrials such as GE Vernova, Caterpillar and Comfort Systems, which had roughly doubled, have been weak since May. Small caps and rate-sensitive sectors such as utilities and real estate have also done badly. Big tech was doing more and more of the work, and it cracked last week as yields rose. He takes that as a late-cycle pattern consistent with higher bond yields pressuring equities. He says the euphoria is broken and sentiment is more wary, but he is not predicting a crash.

  “we have seen the market behaving in a way that is consistent with the end of a cycle”at 25:01

Market sentiment, positioning & volatility

• Robert Armstrong · Financial Times

  Investors are fleeing into consumer staples as fear rises, with staples the worst performers until recently.

  Armstrong says US consumer staples such as paper towels, deodorant, tomato sauce and tobacco had been the worst-performing stocks in global markets, partly due to slow growth. He says staples have now surged in the last week or two as nobody wants risk, which he reads as fear at work. He points to investors swapping tech stocks for consumer-staples names like Kimberly-Clark.

  “Nobody wants anything to do with these very safe, boring companies... So that's fear at work.”at 28:56