Nvidia (NVDA), Microsoft (MSFT) and Anthropic recycle billions to prop up sky-high AI valuations
The AI juggernauts are tightening their grip, with Nvidia (NVDA), Microsoft (MSFT), and Anthropic unveiling a new strategic partnership that further cements their dominance.
However, beneath the celebratory press releases is a more troubling trend: an increasingly circular AI economy where the same handful of companies are funding, supplying, and buying from each other.
According to the terms of the new arrangement, Anthropic has agreed to purchase $30 billion worth of Azure compute capacity.
Nvidia will invest $10 billion in Anthropic, while Microsoft will commit another $5 billion. Nvidia and Anthropic will also collaborate on chip and model design as part of what they call a “deep technology partnership.”
On paper, it's a virtuous cycle. In practice, it raises questions about whether AI’s biggest players are creating a self-reinforcing system where capital circulates between a few dominant firms, inflating valuations, concentrating power, and crowding out emerging competitors.
Deals of this scale blur the line between genuine market demand and strategic buying intended to justify continued hypergrowth.
This dynamic is exactly what the Yale School of Management warned about in October. In an analysis by Jeffrey Sonnenfeld and co-author Stephan Henriques, the pair argued that “the tangle of AI deals among tech giants could be signs of dangerous overinvestment in the developing technology.”
They cautioned that such intertwined, high-stakes arrangements could ultimately set the stage for an AI bubble — and a painful correction if real-world demand fails to match the industry’s soaring expectations.
The hidden costs behind the AI boom
Some of these concerns have recently been echoed by famed investor Michael Burry, best known for betting against the subprime mortgage market ahead of the 2008 financial crisis.
Burry has cautioned that the AI narrative could unravel in part because of an overlooked accounting issue with major implications: depreciation.
As InvestorsObserver recently reported, several Big Tech companies have quietly extended the assumed useful lifespan of their AI hardware. Doing so allows them to reduce annual depreciation expenses and report higher profits — at least on paper.
The warning comes amid broader fears that AI investment levels may not be justified by current economic returns.
A recent analysis found that U.S. AI data centers are depreciating at a rate of roughly $40 billion per year and would need to generate about $480 billion in annual revenue to deliver normal returns.
Instead, they’re earning only a fraction of that amount, prompting analysts to liken the AI build-out to a “mini–war–time economy” driven more by strategic urgency than sustainability.
Naturally, Nvidia’s growth story sits at the center of these concerns, given that it supplies the core hardware powering the entire AI boom.
If the underlying economics of AI infrastructure turn out to be weaker than expected, Nvidia’s outsized role makes it especially vulnerable to a broader reset in the sector.