Two Companies Are Carrying the AI Boom. They're Losing 59 Cents on the Dollar.

The most profitable companies in the artificial intelligence economy do not build models. They sell silicon and equipment. Their margins run at 41%. The companies that actually build the models — the ones whose technology the entire boom is organized around — operate at negative 59% margins. ^[1]

The bottom line is that the most profitable part of the AI value chain depends on the least profitable part continuing to grow revenue or raise capital. — Torsten Slok

That asymmetry, identified by Apollo's chief economist Torsten Slok, is not a temporary imbalance that scale will resolve. It is the structural condition of the AI capital cycle: the upstream profits depend on the downstream losses continuing. And the downstream is concentrated in two companies. Those two startups are OpenAI and Anthropic. Together they account for roughly 70% of AI-related revenue and between a quarter and a third of cloud revenue at Microsoft, Amazon, Google, and Oracle, according to investor Steve Eisman. ^[2] The majority of the more than $700 billion committed across chips, data centers, and model development flows through them — making them the load-bearing counterparties for an investment cycle that now rivals the build-out of the internet. Follow the money around the circle and the concentration sharpens at each node. Investor capital flows into OpenAI and Anthropic, which lose 59 cents on every dollar of revenue. Those losses are not waste; they are payments to cloud providers. OpenAI has a $300 billion five-year compute deal with Oracle — a contract that alone exceeds half of Oracle's total backlog. ^[3] It also holds a $250 billion Azure contract with Microsoft, which owns a 27% equity stake. ^[3] Anthropic has a rumored $200 billion five-year commitment that forms the backbone of Google Cloud's $460 billion backlog. ^[4] The cloud providers, in turn, buy hardware from Nvidia. Nvidia then invests back into OpenAI. ^[5]

Excited to partner with AMD to use their chips to serve our users! — OpenAI

The cloud revenue growth that appears to validate the boom — Google Cloud up 82%, Azure up 43% and exceeding $100 billion, AWS up 37% — is itself substantially a pass-through. ^[6] Eisman's estimate that OpenAI and Anthropic represent 25% to 35% of total cloud revenue for the four hyperscalers captures their share of the existing revenue base. But AI demand is the marginal driver of cloud growth, and these two companies are the dominant AI spenders. Much of the cloud revenue surge is the same money, counted at a different node in the circle. The pressure is building at the revenue end. Chinese competitors led by DeepSeek have triggered a price war: OpenAI cut fees by 80% on its Luna model, and Anthropic introduced a high-performance model at half the cost of its top system. ^[7] DeepSeek's V4-Flash costs roughly three cents per benchmark test — more than 100 times cheaper than Anthropic's Claude Fable 5 — and Bloomberg Intelligence warns the sector may not reach profitability for another three years. ^[8] Meanwhile, an MIT study found that 95% of generative AI pilots at 52 organizations generated zero return on investment. ^[9] A Bank of England survey found that 90% of senior managers report no measurable impact on labor productivity from AI. ^[10] The industry has invested roughly $717 billion over three years, while total revenue across OpenAI, Google, and Anthropic grew from $1 billion in 2023 to $4 billion in 2024 — a 180-to-1 spending-to-revenue gap. ^[11] The cascade, if it comes, runs in reverse. If either OpenAI or Anthropic cannot raise capital — because the revenue trajectory breaks, or because investors lose patience with a sector where 95% of pilots produce nothing — their cloud contracts shrink. Cloud backlogs shrink. Hardware orders fall. And here the fragility extends beyond the tech sector. GPUs are increasingly used as collateral in private credit vehicles. ^[1]

instead of IPOs, Nvidia is the IPO, funding everyone and anyone. — Cost Plus Drugs

If AI demand weakens and GPU values fall, distress propagates to banks, insurers, and pension funds. The Bank for International Settlements — the central bank for central banks — has warned explicitly that disappointment in AI returns could trigger a sudden pullback in financing and turn the capital-expenditure boom into a protracted investment bust. ^[1]

Disappointment in returns could trigger a sudden pullback in financing and turn the capex boom into a protracted investment bust, with potential knock-on effects on financial conditions. — Bank for International Settlements

There is a bull case, and it is not trivial. OpenAI projects $100 billion in annual recurring revenue by 2027, a year ahead of schedule, and $200 billion by 2030. ^[12] BlackRock estimates $5 trillion to $8 trillion in global AI spending by 2030. ^[13] Amazon CEO Andy Jassy says equipment purchases break even in under three years. ^[6] And DeepSeek itself is now raising prices ahead of a potential IPO, suggesting the price war may moderate. ^[14] But the bull case and the bear case describe the same structure. The $100 billion revenue projection, the $5 trillion spending forecast, the cloud growth numbers — all of them are aggregate figures that obscure how much runs through two entities. The boom looks diversified because every node has a counterparty: cloud providers, chipmakers, equipment lessors, model labs, enterprise platforms. But at the dominant junctions, the counterparty traces back to the same two companies burning investor capital. The market reads breadth where there is concentration. And the people inside the structure know it. Jeff Bezos called it "kind of an industrial bubble." Sam Altman acknowledged that "people will overinvest and lose money." ^[9] They are not outside critics. They are the counterparties.