Who is really paying for the AI boom? Follow the money from cloud to AI labs

The artificial intelligence economy is beginning to look less like a simple supplier-customer relationship and more like a financial loop.

Microsoft, Amazon and Google are spending enormous sums to build the computing infrastructure needed for AI. At the same time, the biggest AI laboratories are becoming major customers of those very cloud platforms, creating a powerful cycle of spending, revenue and reinvestment, according to a report by Jefferies’ Global Head of Equity Strategy Christopher Wood.

Jefferies' latest report highlights just how significant that relationship has become.

The combined cloud revenues of Microsoft, Google and Amazon reached $126 billion in the second quarter of 2026, up 38% year-on-year. But behind those numbers sits a rapidly growing dependence on AI customers.

MUST READ | AI boom's hidden bill: Big Tech's spending hits $165 billion as free cash flow shrinks

Jefferies estimates that OpenAI and Anthropic together accounted for roughly 6% of Amazon Web Services' total cloud revenue in Q2 2026, including legacy non-AI services. The firm expects their share to rise to 12-18% in Q3.

The dependence is even more pronounced at rival cloud platforms. The two frontier AI labs were estimated to account for roughly 45% of Google Cloud Platform revenue and 25% of Microsoft's Azure revenue in Q2 2026.

That raises a fascinating question: is the AI industry creating genuinely new economic activity, or is a significant portion of the money circulating within the same ecosystem?

The $2.34 trillion backlog  

The scale becomes clearer when looking at contracted future revenue.

The combined remaining performance obligations, or RPOs, of the major hyperscalers reached $2.34 trillion at the end of Q2 2026, up 186% from $816 billion a year earlier.

DON'T MISS | The AI reality check: 85% of hyperscalers' cloud revenue is still non-AI

Jefferies estimates that the top two frontier AI labs account for roughly 45-50% of Amazon's backlog, 40% of Google's and 30-40% of Microsoft's.

In other words, some of the future cloud revenue being used to justify today's enormous AI infrastructure investments is tied to AI companies themselves.

That does not automatically make the model unsustainable. AI labs are generating rapidly growing revenues and are increasingly monetising their products. Jefferies cites industry estimates putting the combined annualised revenue run rate of OpenAI and Anthropic at around $115 billion.

DO CHECKOUT | Meta's Mark Zuckerberg says AI creates jobs as data centres expands

But the structure does create a new risk to watch.

If AI labs continue to grow rapidly, hyperscalers can justify ever-larger infrastructure investments. If AI demand slows, however, the same ecosystem could face pressure from both sides: AI companies could reduce infrastructure consumption while hyperscalers are left carrying enormous capital costs.