Michael Burry Called Nvidia’s $95 Billion in Purchase Obligations ‘Troubling’ in February. Six Months Later, the Filings Show That Number Ballooned to $279 Billion.

Michael Burry, the investor made famous by “The Big Short,” told readers of his Substack in February that there was “one specific aspect of their financials that I find troubling” about Nvidia (NVDA). The number he pointed at was purchase obligations of $95.2 billion in the company's fiscal 2026 annual report, “up from $16.1 billion the same time last year.” Six months later, Nvidia has filed again, and the comparable commitment line is far larger.

The Feb. 26, 2026 post was titled “Short Thought: Nvidia Ratchets Up the Risk.” Burry's summary of what the figure meant, quoted at the time by CNBC and Business Insider, was blunt: “This is not business as usual. This is risk.” He wasn't describing a crash in progress, just a balance sheet he thought had taken on a new kind of exposure, and Barchart reported the post when it landed.

A purchase obligation is a promise to buy. When a chip designer reserves wafer capacity, advanced packaging slots, and high-bandwidth memory years ahead of shipment, it signs contracts that are largely non-cancelable. Orders are placed before anyone knows whether the demand at the far end will still be there. On the way up, that is how a company guarantees itself a supply nobody else can get. On the way down, it is a bill that arrives regardless.

Nvidia's quarterly report for the period ended July 26, 2026, puts long-term supply commitments at roughly $279 billion, against about $119 billion three months earlier, an increase of some $160 billion in a single quarter, which the company attributes primarily to the procurement of memory. The maturity ladder in the same filing shows roughly $92 billion falling due across the remainder of this fiscal year and about $87 billion in fiscal 2028.

Sponsored Content: This Startup Is Targeting a Massive Untapped Energy Market

One caution on the arithmetic, because it is the easiest thing to get wrong here. Burry cited “purchase obligations” of $95.2 billion from the annual report, alongside a broader total supply obligation figure of roughly $117 billion in the same post. The August document reports “long-term supply commitments.” Those are not guaranteed to be the identical line item, so the cleanest like-for-like comparison is the one inside the quarterly filings themselves: about $119 billion in April, and about $279 billion in July.

What made Burry's post travel was the comparison he reached for. He set Nvidia beside Cisco Systems (CSCO) in the months before the dot-com collapse. A networking company that could not keep up with orders ended up writing off a large share of the inventory and commitments it had raced to secure. Burry's conclusion was a conditional, and the conditional here matters: “Any downturn, when it comes, will be more severe, perhaps even catastrophic, for Nvidia's earnings and balance sheet.” He said if, not when, it would happen.

Nvidia's answer arrived on the same day as the filing. Chief Financial Officer Colette Kress addressed the criticism directly on the Aug. 26 earnings call, saying of the company's support for its supply chain and customers: “We recognize the scale of this support, and we know some will call this circular financing. We see it differently.” The company's position is that reserved capacity is a competitive asset rather than a liability and that the components underneath it are fungible across a growing product line.

The revenue line is what separates this from a straightforward bear case. Nvidia reported second-quarter revenue of $96.2 billion, up 106% from a year earlier, and guided the current quarter to roughly $108 billion. On the same day the $279 billion figure became public, Amazon (AMZN) disclosed a sharply increased order for Nvidia silicon, and Chief Executive Jensen Huang forecast fiscal 2028 revenue growth well above what analysts had modeled. Huang has spent months arguing the opposite of Burry's case, telling audiences there is no AI bubble because AI is “incredibly profitable.”

So the same disclosure supports two readings. One says a company has pre-bought the entire world's memory supply because it can see demand nobody else can. The other says a company has committed a quarter of a trillion dollars against demand that is currently being forecast rather than delivered. Cisco's revenue was decelerating when its commitments came due. Nvidia's is doubling.

The thing that settles it is not rhetoric on either side. It is whether the revenue line keeps growing faster than the commitment line over the next two quarters and whether the $92 billion due this fiscal year converts into shipped product or into inventory. Both numbers are disclosed. Both get updated in 90 days.