If the Labs Wobble, the Clouds Feel It First
I have written separately about the shaky economics behind the data center boom and the growing pressure on the frontier labs. This post connects the two. Microsoft reported $24.1 billion of revenue from its OpenAI relationship, an amount equivalent to almost one-quarter of Azure’s scale, although…
I have written separately about the shaky economics behind the data center boom and the growing pressure on the frontier labs. This post connects the two. Microsoft reported $24.1 billion of revenue from its OpenAI relationship, an amount equivalent to almost one-quarter of Azure’s scale, although the figure includes revenue-sharing payments as well as cloud consumption. The best estimates put OpenAI and Anthropic at another 11% to 14% of AWS. More important, Microsoft’s commercial backlog grew 84%, but only 25% without OpenAI. That does not prove an AI bubble, but it does show how concentrated the next stage of cloud growth has become. A large part of the infrastructure boom is being built today against future spending commitments from a handful of private companies whose revenues, margins, and funding needs remain difficult to assess. AI can keep improving while the assets financed around it still produce disappointing returns. That concentration would worry me less if OpenAI and Anthropic had obvious, durable pricing power. I do not think they do. Open models keep narrowing the gap, customers are routing more work to cheaper systems, and enterprises can increasingly turn their own data and workflows into specialized models they control. The labs are not only competing with one another. They are beginning to compete with the capabilities their largest customers can build for themselves. If lab spending slows, the pressure moves quickly into cloud growth, data center utilization, and the financing arranged around future demand. Oracle is the case I would watch most closely. Its current OpenAI revenue appears modest precisely because the larger revenue ramp has not arrived, while the company has already committed tens of billions to infrastructure and financing. For buyers signing long-term compute agreements and investors underwriting the buildout, the useful question is no longer just whether AI succeeds. It is who ultimately has to keep paying for the capacity, and how strong the business on the other side of that promise really is. Subscribe to our weekly newsletter Appendix: a word on the estimates The figures in the graphic should not be treated as equally precise. Microsoft’s $24.1 billion is the only disclosed customer-level number, and it includes revenue-sharing payments, so it measures the broader OpenAI relationship rather than pure Azure usage. The AWS range is the strongest estimate, with Barclays and Wolfe Research landing close together on Anthropic but differing estimates for OpenAI. Google’s roughly 7% combined figure comes from a single January estimate that may now be low, while a more aggressive UBS analysis cannot be used as an upper bound because it also includes Meta and lower-margin TPU hardware sales. Oracle’s range is not a published current-year estimate. It is a working inference based on OCI’s reported growth and analyst projections that place the larger OpenAI revenue ramp in later fiscal years. All percentages compare the labs with the cloud platforms, not with total corporate revenue. The post If the Labs Wobble, the Clouds Feel It First appeared first on Gradient Flow.Source: Gradient Flow — Published — Category: Models