Google Cloud Revenue Reliance on OpenAI and Anthropic

Google Cloud revenue reliance on OpenAI and Anthropic could become a defining risk in the next phase of AI infrastructure spending. Ed Zitron, chief executive of EZ Primary Research, argued in a Bloomberg interview that hyperscaler capital expenditure is creating capacity for two loss-making customers, while analyst estimates imply that those customers could represent an unusually large share of future cloud revenue. For builders, the practical concern is simple: cloud pricing and capacity plans may be shaped by financing decisions outside the cloud business itself.

The estimates point to an unusually concentrated cloud customer base

UBS estimates cited by Zitron project OpenAI and Anthropic together generating more than $124 billion of Google Cloud revenue in 2027, with Anthropic contributing about $76 billion. The same analysis says the pair would account for 27% of Google Cloud revenue this year and more than 48% next year. These are analyst estimates reported through the interview, not audited Google Cloud customer disclosures.

The comparison with Microsoft is also significant, but it should not be blended with the Google figures. Zitron cited Barclays estimates placing OpenAI and Anthropic at 13% of Microsoft Intelligent Cloud revenue this year and 18% next year. He separately claimed OpenAI drove 69% of that segment’s year over year growth in 2025. That claim suggests dependence on a single customer, although the supplied evidence does not independently verify the calculation.

Why this matters to teams building on cloud APIs

A cloud provider can tolerate a large customer when that customer has durable cash flow and diversified demand. The concern raised here is different. Zitron says OpenAI lost $20.9 billion in 2025 and relies on continuing capital inflows rather than existing operating cash to pay its bills. If either model company slows spending, loses access to financing, or changes providers, the impact could reach data center expansion and capacity planning.

That does not mean a sudden cloud outage is the likely result. The more practical risk for an AI product team is slower capacity growth, revised pricing, tighter quotas, or less predictable discounts. Teams running long agent workloads should model those possibilities instead of assuming that today’s token prices and reserved capacity offers will remain unchanged.

The ad-funded infrastructure loop is the important part

The analysis connects AI infrastructure capital expenditure to the advertising businesses operated by Alphabet, Microsoft, Amazon, and Meta. Those businesses generate cash that can help fund data centers, while the same infrastructure supports ad targeting, creative generation, programmatic bidding, search features, and shopping assistants.

This creates a feedback loop, but not proof of an unsustainable market. Advertising revenue can fund productive infrastructure if AI services generate enough durable demand. The unresolved question is whether demand comes from a broad base of customers or from a small number of model companies whose own spending depends on external investment.

A sensible response is diversification, not panic

For builders, vendor diversification is now a capacity and finance decision, not only a reliability exercise. Keep model routing portable, test at least one alternative provider, track rate limits and regional availability, and separate application logic from provider specific features. Finance teams should also include migration work, fallback inference, and higher burst pricing in total cost of ownership models.

The evidence supplied here supports a concentration warning, not a definitive prediction that Google Cloud economics will fail. The figures come from an interview and estimates cited by EZ Primary Research, with no independent corroboration provided in the research pack. The decision rule for builders is therefore practical