Microsoft, Google and Amazon are touting explosive AI growth, but a new wave of analyst research suggests that boom is largely coming from just two customers: OpenAI and Anthropic, both of which are still burning enormous piles of cash to keep their models running.
Tech writer Ed Zitron pulled the numbers together in his newsletter Where’s Your Ed At, arguing that an “AI demand bubble” is being propped up by eye‑watering spending from those labs rather than broad customer adoption of cloud AI. Zitron, drawing on research from Wells Fargo, Barclays, UBS and Deutsche Bank, estimates that roughly 74% of Microsoft’s AI revenue, 73% of AWS’s AI revenue and more than 70% of Google Cloud’s AI revenue are tied to OpenAI and Anthropic’s compute bills. The banks also peg OpenAI and Anthropic’s share of total cloud revenue at 23% of Microsoft Azure’s take in fiscal 2026, rising to around 35% in 2027, with similarly outsized shares projected for Amazon and Google.
Microsoft’s own filings back up just how concentrated that business has become, disclosing about $24.1 billion in revenue from OpenAI in the year ending June 2026—roughly 70% of the company’s entire AI sales for the period. Separate reporting notes that OpenAI commitments now make up roughly 45% of Microsoft’s $625 billion cloud revenue backlog, meaning nearly half of the AI and cloud dollars Microsoft expects to earn in coming years are already locked in with a single partner. That entanglement sits on top of Microsoft’s multibillion‑dollar equity stake in OpenAI and the company’s role as the startup’s primary infrastructure provider and biggest downstream customer.
In April 2026, Microsoft and OpenAI quietly rewrote the terms of their alliance, ending Microsoft’s obligation to share a cut of its AI revenue with OpenAI and instead locking in a capped revenue‑share stream from OpenAI back to Microsoft through 2030. Under the revised deal, OpenAI continues to hand over around 20% of its revenue to Microsoft, but total payments are now capped at roughly $38 billion, down dramatically from earlier projections that stretched as high as $135 billion. Documents obtained by Zitron also detail how Microsoft invoices OpenAI for massive Azure inference workloads while simultaneously skimming 20% of the revenue OpenAI earns from products like ChatGPT and its API platform.
All of this is happening while OpenAI and Anthropic remain deeply unprofitable, plowing revenue straight back into GPU time and data‑center capacity so they can keep cranking out ever‑larger foundation models. The result is a strange ouroboros: AI labs spend huge sums on cloud compute, that spend shows up as “AI revenue” for Azure, AWS and Google Cloud, and those same numbers are then waved at investors as proof that generative AI is already a massive, mainstream business. If those two labs pull back on spending—even without an AI winter—the headline growth numbers for Big Tech’s AI segments could fall off a cliff.
Amazon and Google each have their own strategic hooks into Anthropic, with AWS and Google Cloud racing to be the default home for the company’s Claude models even as they lean on Anthropic’s heavy usage to pad their AI revenue stories. For users who only see the front end—Copilot baked into Windows, Gemini inside Chrome, Claude chatting in Slack—this backend reality means the future of their favorite AI tools is tightly bound up with the financial health of a tiny handful of labs and the willingness of cloud giants to keep subsidizing their experiment. Investors, regulators and everyday geeks now have to ask whether the AI boom is a durable new platform or just another bubble inflated by a couple of money‑losing companies buying compute at a historic scale.
