Nvidia’s Chip Trail Comes Back as a Disclosure Problem

Nvidia received a subpoena from California’s attorney-general on Nov. 19, which converts a dispute about where its processors end up into a question about what the company told investors about them.

The same day, Democrats on a House committee wrote to the company demanding records tied to the inquiry into Megaspeed, a Singapore-based supplier that resold capacity on Nvidia hardware. One track is state, one is federal, and both end at the same document set: purchase orders, customer lists, internal assessments of who is on the other end.

The mechanism is worth spelling out, because it explains why the two tracks exist. Nvidia books revenue by the location of the entity that signs the invoice, not by the address of the data centre where the hardware is installed. Those two facts are frequently different. A distributor in Singapore can bill a customer in Manila for a rack that serves users in Guangdong, and every party in that chain can answer a questionnaire truthfully while the hardware sits somewhere none of them named.

That gap became visible in 2025, when Singapore-billed revenue rose from 8% to 22% of the total and the company stopped disclosing the split. Nvidia has said the figure reflects where its customers are headquartered or billed rather than where chips are used, which can be entirely true and still leave a regulator without the answer he needs.

Export licensing has always rested on the buyer’s paperwork. Washington’s Bureau of Industry and Security grants or denies licences on end-user declarations, and the enforcement cases that follow are built from shipping manifests, freight forwarders and whistleblowers rather than from a chipmaker’s general ledger. A subpoena from a state attorney-general is unusual because it aims at the ledger.

California’s top prosecutor holds broad subpoena power under the state’s unfair-competition statute, a tool that has been used against conduct federal agencies chose not to pursue. Investor-protection theories built on state law have become a practical route to documents a company would rather discuss in a federal forum with a cooperative regulator.

The House letter widens the audience. Lawmakers have been asking since 2025 how advanced chips reached Chinese users through third-party clouds, and a formal records request turns press reporting into a docket. Material produced to Congress has a way of becoming public, which changes the calculation for every customer named in it.

Then there is the filing calendar. Nvidia delayed its 10-Q, citing executive changes. Audit committees rarely treat management turnover as a reason to postpone a quarterly report, and at a company of this size a delay reads as a question about internal controls rather than about paperwork.

Assembled, the exposure is not mainly about the chips. Export controls are a licensing problem that can be cured with a permit, a divestiture or an inventory write-off, as the H20 episode showed when Nvidia took a $4.5 billion charge and later accepted a revenue-sharing arrangement to resume sales to China. Disclosure is harder to cure, because it turns on what was known and when it was known.

The China question has already been repriced by the market. Nvidia’s data-centre revenue from China fell from about a fifth of the total to a rounding error in three years, and its valuation now rests on American, Gulf and European demand. A finding that some of that demand was routed through intermediaries would hurt the multiple less than a finding that the company’s own statements about it were wrong.

There is a precedent that maps closely. Seagate paid $300 million in 2023 to settle US charges that it shipped hard drives to Huawei after the restrictions took effect, a case built on internal emails about which customer would end up with the equipment. The product was mundane and the penalty was not. The lesson was that “we did not know” is a defence that requires evidence.

For customers, the subpoena changes the arithmetic of buying Nvidia. Hyperscalers and rented-compute providers now have to assume their names, order sizes and locations will be read by investigators, and some will prefer suppliers whose paperwork produces fewer questions. That is a market-share risk with nothing to do with performance per watt.

For the industry, the lesson is that the chip is no longer the unit of control. Compute is rented, resold and reached remotely, and a rack in Singapore can serve a user in Shenzhen without a component crossing a border. Rules written for boxes on ships do not reach that arrangement, and every large buyer knows it, which is why the compliance function has grown faster than the sales function at several suppliers.

The question the subpoena poses is narrow and awkward. Can a company with tens of thousands of customers and a three-tier distribution chain know, and then honestly disclose, where its product is used? If the answer is no, the disclosure standard will have to be written down to something a chipmaker can actually certify, and investors will have to price the difference between what is true and what is attestable.

Related Posts

  • November 20, 2026
  • 15 views
Sandisk Completes Eight-for-One Split as NAND Outlook Darkens

Eight for one. Sandisk’s stock split took effect on Nov. 19, and the shares began trading on the adjusted basis on Nov. 20, a change that divides the price by…

  • November 20, 2026
  • 13 views
A Lawmaker’s Privacy Suit Puts Model Design on Trial

For three years British courts have sidestepped the question of whether the company that builds an AI model answers for what its users produce with it. A written defence filed…