The headline Baker wants investors to see
Gavin Baker’s appearance on TBPN landed at a moment when Anthropic’s newly public S-1 had given investors an unusually detailed look inside one of the defining AI companies of the cycle. Anthropic announced on October 26, 2026 that it had filed a registration statement with the SEC for a proposed initial public offering of Class A common stock, with plans to list on the Nasdaq Global Select Market under the ticker “ANT,” while the share count and price range remained undetermined . The SEC filing itself was dated October 26, 2026, and described a preliminary prospectus whose terms were still incomplete and not yet effective .
Baker’s interpretation was emphatically bullish, but not simplistic. His core claim was that Anthropic’s S-1 was “pretty clearly an upside surprise,” because the disclosures showed stronger profitability, cash generation and unit economics than skeptics expected . In his reading, the filing did not describe an AI lab merely burning capital to buy revenue; it described a company whose demand is already broad, whose customer base is less concentrated than feared, and whose enterprise penetration still looks early.
The most important number for Baker was not only revenue. It was adoption relative to corporate spend. Anthropic’s S-1 says that, as of September 30, 2026, it had more than 11,900 customers with at least $100,000 of trailing-12-month revenue, more than 1,900 customers above $1 million, and about 250 customers above $10 million . It also says that over 50% of the Fortune 500 and over 30% of the Global 2000 were customers above that $100,000 threshold, while revenue from those Fortune 500 customers averaged less than 0.1% of their total expenses in the third quarter of 2026 .
Why “0.1% of expenses” matters
That 0.1% figure is the hinge of Baker’s argument. If more than half of the Fortune 500 is already using Anthropic, but the average spend is still a rounding error in those companies’ cost structures, then AI adoption is not close to saturation. It is early. Baker put the point sharply: getting to that level of Fortune 500 adoption within a few years is impressive, but the bigger question is what happens when the other half adopts and the first half moves beyond pilot-scale or departmental usage .
The S-1 supports that “early but broad” framing in several ways. Anthropic says no single customer represented more than 10% of revenue for the nine months ended September 30, 2026, and that about 79% of third-quarter revenue came from customers outside its top five, up from about 64% in the March quarter . That undercuts a simple bear case in which one or two hyperscale customers are imagined to be propping up the entire number.
At the same time, the filing shows just how fast the revenue base has expanded. Anthropic reported $16.286 billion of revenue for the six months ended June 30, 2026, up from $1.163 billion in the comparable 2025 period, a 1,300% increase . Compute and infrastructure expense also rose sharply, to $8.945 billion in the first half of 2026, but fell as a percentage of revenue from 201% to 55% year over year . That is the backdrop for Baker’s view that the public numbers look healthier than many expected.
GPU prices are the stress test
The twist in Baker’s thesis is that better unit economics do not remove the bottleneck. They intensify it. In the TBPN discussion, he said October had been unusual because the rental price for “essentially any kind of GPU” had gone up every day, suggesting that the spot compute market was tightening rather than easing . He then extended the point from spot pricing to the forward curve: according to Baker, Silicon Data founder Carmen Li said at GTC that the compute market had moved into contango, meaning future capacity was pricing above current capacity .
That matters because investors keep asking whether the announced wave of GPU supply will overwhelm demand in 2026 and 2027. Baker’s answer is that demand is already chasing that supply. He described portfolio companies that had secured 4,000 Blackwell GPUs for May or June 2026, only to be surprised by how much more expensive another 4,000-GPU cluster looked for September 2027 . In a separate summary of the same discussion, Baker’s point was framed as a warning that future cluster costs are rising even as new capacity is planned .
This is why the S-1 and GPU pricing belong in the same story. Anthropic’s filing describes compute as the company’s most significant operating cost and says its ability to generate more revenue per dollar of compute depends on training efficiency, inference efficiency and a multi-chip architecture spanning AWS Trainium, Google TPUs, Nvidia GPUs and AMD GPUs . In plain English: if AI revenue keeps scaling, compute does not become an afterthought. It becomes the scarce production input.
SpaceX and the operator premium
SpaceX enters Baker’s framework not as a side note, but as an example of execution becoming part of the economics. The issue is not just who can order GPUs. It is who can finance them, energize them, connect them to power, keep them running and choose contract structures that preserve upside when market prices move .
Baker praised CoreWeave as a strong GPU operator while arguing that some of its long-term contracts were fixed at prices now below market . That contrast led the TBPN hosts to ask whether Elon Musk and SpaceX had studied that lesson and chosen a shorter-duration, more flexible approach . Baker’s broader point was that running GPU clusters is a physical-world execution problem: fast deployment and reliable operation can determine whether expensive hardware becomes revenue quickly or sits idle.
Orply’s account of the interview emphasized the same operating theme: Baker argued that new compute supply does not automatically mean cheaper compute, because the market must still absorb the cost and complexity of bringing clusters online . He also said public listings by Anthropic and SpaceX could free up buying power for private-market AI investments, particularly among mutual funds constrained by liquidity rules . In that sense, the IPO cycle and the compute cycle are linked: public exits can recycle capital into the next infrastructure buildout.
The bear case Baker still respects
Baker is not presenting the story as risk-free. He described himself as a “responsible optimist” and singled out one bearish argument he found compelling: if a large share of Anthropic’s compute is devoted to training, then any shock that reduces training demand could push capacity into the inference market and pressure pricing . That is the right caveat. The same compute scarcity that supports today’s economics could reverse if model-development assumptions change quickly.
Still, Baker’s October read is clear: Anthropic’s S-1 makes enterprise demand look broader and earlier than expected, while GPU markets suggest that the infrastructure bottleneck has not been solved by supply announcements alone . The surprising conclusion is that falling token prices and rising GPU rental prices can coexist. If cheaper AI unlocks more usage, the demand for compute can rise even as model-layer pricing compresses.
The “lightspeed” joke around the GPU bottleneck works because it captures the contradiction. The market keeps hearing about more chips, more power and more data centers. Baker is saying that the bottleneck has not disappeared; it has moved forward in time, into 2027 contracts, operator quality, financing structures and the physical speed of deployment. Anthropic’s S-1 gave investors the demand signal. GPU prices, and SpaceX’s role in the buildout, show where the next constraint may bite.
Comments
Be the first to comment.