The AI boom has made Nvidia the stock market’s darling – and its favourite scapegoat. With a share price that has decoupled from historical norms, many analysts treat the chipmaker as the inevitable centre of the next great market collapse.
Yet to assume Nvidia is the bubble is to fundamentally misunderstand its role. Nvidia sells the essential tools everyone else needs; it does not sell the large language models or subscription services that rely on unpredictable user behaviour. The true bubble risk lies with the commercial software firms betting that AI will eventually deliver the revenues their valuations demand.
Nvidia Is a Supplier
Nvidia builds the technology rather than the products that sit on top of it, which puts it in a different category from the companies trying to turn AI into a commercial service.
There is no shortage of candidates carrying the real risk. Palantir, for example, currently trades at a trailing price-to-earnings ratio above 200 times – a valuation that only makes sense if AI becomes the indispensable centre of corporate decision-making worldwide. Smaller AI software names tell similar stories, with market caps that have surged despite the absence of any proven evidence that customers will pay for these tools at scale.
The hope is that AI will become the next must-have productivity engine. The problem is that hope remains the primary piece of evidence on offer.
The Circular Economy
This reliance on speculation is why critics increasingly point to what they describe as a recursive closed-loop economy – and the deals that have emerged in the past eighteen months make the concern concrete.
In September 2025, Nvidia agreed to invest up to $100 billion in OpenAI, tied to OpenAI committing to deploy at least ten gigawatts of data centre capacity using Nvidia systems. The chip vendor was effectively bankrolling its own future sales. Critics were quick to flag the circularity: the same dollar leaving Nvidia’s balance sheet was guaranteed to return as hardware revenue.
That deal is not an isolated case, with money, chips and cloud credits now rotating in a closed loop among a small group of companies including Nvidia, OpenAI, Microsoft, Oracle, AMD and CoreWeave, where equity stakes, compute commitments and capacity guarantees flow in multiple directions at once, creating the appearance of breakneck revenue growth even though part of it is effectively the same capital circulating under different guises.
The Accounting Question
This fragile dynamic is further obscured by a shift in how major cloud providers account for their hardware.
Amazon, Alphabet, and Microsoft have all extended the official useful life of their servers from three or four years to six – a coordinated move that meaningfully flatters reported profits by spreading the same capital cost over twice as many years. Amazon’s most recent extension alone added roughly $3.2 billion to its operating income in 2024. As AI infrastructure spending accelerates into the hundreds of billions, the gap between reported earnings and cash reality widens considerably.
Investor Michael Burry has argued that AI hardware actually becomes economically obsolete within two to three years, given the pace at which Nvidia ships new architectures. If he is correct, the industry may be understating depreciation by an estimated $176 billion between 2026 and 2028 – a reckoning that would hit profits hard when it arrives.
Non-cash arrangements also play a role, with deals struck in computing credits rather than cash making profitability appear stronger than the underlying economics support.
Where the Pressure Lives
If AI software firms struggle to demonstrate genuine profit, investors will not punish Nvidia first. They will punish the companies that promised too much.
A drop in confidence would hit the firms relying on AI hype to justify their valuations. Once that unwinds, the sell-off spreads. Corporate demand for GPUs slows. Capital expenditure plans get pushed back. Nvidia becomes a casualty not because it was a bubble in its own right, but because its customers were.
Right now, there is no clear proof that AI is sustainably profitable beyond the infrastructure layer. Cloud providers make money selling access to Nvidia’s hardware, but the AI products running on top of that hardware have not yet delivered the margins needed to justify the capital poured into them. Even the biggest names in the space have yet to prove that AI revenue can support the enormous cost of running these systems at scale.
The question is not whether Nvidia can cause a bubble, as the pressure points sit further up the chain with the firms selling AI dreams rather than AI results, and if those expectations fade, Nvidia would be caught in the slowdown that follows, though it is not the company inflating the balloon, only the largest player standing closest to it.
