Michael Burry Warns of Customer Concentration Risks at Nvidia

Michael Burry, the investor who became a household name by betting against the US housing market before the 2008 financial crisis, is now turning his attention to what he sees as a glaring vulnerability in one of the world’s most valuable companies. His target: Nvidia’s increasingly lopsided customer base.

In Substack posts dated May 25-26, Burry pointed out that just three customers now represent 64% of Nvidia’s accounts receivable. That figure was 56% only one quarter earlier. Burry called the level of concentration “off the charts,” and he’s already put money behind his conviction by acquiring put options against the chipmaker.

The customers driving this concentration are hyperscale cloud providers, think Microsoft, Google, Amazon, and Meta. These companies have collectively driven more than 50% of Nvidia’s data center revenue in certain reporting periods. Nvidia’s own Q2 2025 SEC filing showed that two unnamed customers accounted for 39% of total revenue, split at 23% and 16% respectively.

Burry also flagged an interesting divergence at Microsoft specifically. Microsoft’s share of Nvidia’s receivables reportedly rose even as its revenue contribution to Nvidia declined for the first time in over three years. In plain English: Microsoft owes Nvidia more money proportionally, but is actually buying less.

Burry is drawing explicit parallels to Cisco Systems during the dot-com era. Cisco was the picks-and-shovels play of the first internet boom. Then the bubble popped. Cisco’s concentrated revenue streams evaporated as telecom companies slashed spending, and the stock fell roughly 80% from its peak. More than two decades later, it still hasn’t recovered to those highs.

Nvidia is the undisputed picks-and-shovels play of the AI boom. Its GPUs power the training and inference workloads that hyperscalers are racing to build out. Burry’s thesis hinges on whether the current AI spending spree is a structural shift or a cyclical surge, and specifically whether the “training phase” he references proves to be a finite period rather than a permanent state.

Back in November 2025, Burry flagged what he described as questionable accounting practices among hyperscalers, specifically that companies were underreporting depreciation on Nvidia GPUs by extending their useful life estimates to 5-6 years. If hyperscalers decide their existing GPU fleets have longer useful lives, they have less incentive to buy new ones. The very accounting trick that makes their current spending look sustainable could end up reducing future demand.

The competitive landscape adds another layer. While Nvidia currently dominates the AI accelerator market, companies like AMD, Intel, and a growing roster of in-house chip efforts from Google, Amazon, and Microsoft are all chipping away at that moat.

Burry has warned of an “aggressive fall” in Nvidia’s share price if hyperscaler spending decelerates. For investors, the key metric to watch going forward is whether that 64% figure continues climbing or starts to moderate.