Inside the AI Bubble: How OpenAI & Nvidia are Weaving a Web of Circular Deals

In recent months, the AI ecosystem has witnessed a string of headline transactions. This includes Nvidia’s multi-year supply and investment commitments to OpenAI’s Stargate datacentre programme, OpenAI’s partnership with AMD to deploy tens of billions of dollars’ worth of its chips to get the chipmaker’s largest shareholding in return, xAI raising $10bn at $200bn valuation, and many more.

The latest deals have collectively created a market narrative, where chipmakers, cloud operators and AI platforms mutually underpin each other’s valuations and financing. Bloomberg framed these arrangements as “circular deals.”

This swirl of mega-deals and a mounting datacentre build-out has led the investors asking whether AI is a boom or a bubble.

An “AI bubble” is the combination of investor euphoria, outsized capital commitments and sky-high valuations that outrun the real, recurring revenue those technologies can reliably generate.

In previous bubbles, like the dotcom and housing bubble, the common pattern was the same- a powerful new narrative attracts capital, that capital chases projects with weak fundamentals and the resulting overhang of debt and excess capacity eventually produces a sharp re-rating.

The current debate around the AI bubble asks whether AI today is at that same juncture, with datacentres, chips and platform deals replacing websites and mortgages.

Circular Deals

Large AI providers need both compute (chips and servers) and data-centre real estate; chipmakers need huge orders to sell accelerators; datacentre builders need long-term tenants and financing.

This interdependence has produced complex commercial relationships. Large customers sign capacity and chip deals that in turn boost chipmakers’ stock and credit capacity, which then underwrites more datacentre and cloud projects.

This “circular” feedback, customers helping vendors, and vendors enabling customers, amplifies growth narratives and, critics argue, can conceal underlying fragility if the real downstream demand doesn’t materialise.

The Spending Spree

The scale of investment is staggering. As per the Guardian’s report, analysts at Morgan Stanley estimate global datacentre spending tied to AI could approach $3 trillion through 2028. The capital will reportedly be utilised to build a massive physical capacity that underpins models, training runs and inference at scale.

Morgan Stanley’s modelling suggests hyperscalers will cover roughly $1.4 trillion of that outlay from their own cashflows, leaving roughly $1.5 trillion of demand to be financed from other sources such as private credit.

These numbers explain why entire cities court datacentre projects and why projects promise tens or hundreds of billions of dollars of new hardware.

Microsoft, Google, Amazon, Meta and others have increased capital expenditure commitments. Morgan Stanley’s forecasts envision generative AI revenues expanding from roughly $45 billion last year toward a potential $1 trillion market by 2028, a revenue case investors are implicitly banking on to absorb this supply.

Huge Valuations: But Is it Worth?

Investors and executives point to eye-popping valuations as a sign of confidence. Nvidia recently crossed a $5 trillion valuation milestone and restructurings and large private rounds have implied very large paper values for AI platform companies.

However, several industry observers warn of classic warning signs. Alibaba’s Joe Tsai said he was starting to “see the beginning of some kind of bubble” in datacentre financing, and analysts at Uptime Institute note many announced sites will either not be built or will be populated slowly.

An MIT research found that 95% of organisations running generative AI pilots were seeing no measurable return from those pilots. If revenues don’t follow the promised curve, assets, from purpose-built AI datacentres to costly GPUs, could sit underutilised while debt providers are left holding risk.

Will the Bubble Burst?

There’s a split. Part of the activity (hyperscaler spending by Microsoft, Google, Amazon and Meta) is arguably “healthy” infrastructure investment backed by enormous cash flows. Another part looks speculative, developers and investors building capacity in expectation of third-party demand that may not materialise.

If enterprise AI monetisation disappoints, or if private credit markets tighten and re-price loans used to fund datacentre construction, losses could cascade beyond tech into lenders and broader credit markets. As one tech analyst put it, the risk is not just to chip companies or cloud vendors but to the providers of the very debt backing the boom.