Who Gets Nvidia’s Chips?
The Huang inquiry puts U.S. export controls under scrutiny.
Nvidia has opened an independent inquiry into whether its chief executive, Jensen Huang, was targeted by a foreign intelligence operation. In Singapore, authorities are separately investigating Megaspeed, an Nvidia customer. Together the inquiries raise a question for U.S. export policy. How much can Washington verify about where advanced AI chips end up and who uses them?
The buyer named on an invoice, the company running the servers and the customer renting computing time can be three different businesses. Each holds a different part of the evidence.
🟥 What does Nvidia’s board need to establish?
Huang acknowledged an affair in a message to employees. Beijing rejects the allegation of Chinese espionage. Nvidia’s board says it has no reason to believe national security or company secrets were compromised. The company’s shares closed more than 4% lower on Monday.
A separate question for Nvidia is whether access to its chief executive influenced orders or compliance decisions. Three U.S. House committee chairs have requested a closed-door briefing on Huang’s contacts with former Megaspeed executive Alice Huang and on Nvidia’s security procedures. The answers depend on who approved the orders and what the people responsible for compliance knew at the time.
🟥 Why does Megaspeed matter?
Megaspeed is a Singapore-based cloud provider and Nvidia customer. On November 10, The Information reported that Alice Huang’s ties to Nvidia and its distributors helped a network of companies supply at least $7.5 billion in chips to Chinese firms. The figure is a reported estimate. Establishing whether particular shipments violated U.S. rules would require evidence about the chips, the dates, the recipients and any licenses.
Singapore says its investigation into Megaspeed continues. It also says it does not enforce other countries’ unilateral export controls and acts against violations of its own laws. The same shipment can therefore raise different questions for authorities in Washington and Singapore.
🟥 Who can verify the delivery address?
U.S. Commerce Department guidance asks companies to scrutinize buyers, end users and intended uses. That scrutiny is difficult when the seller’s records stop at a distributor and a separate business arranges installation. In an unrelated U.S. case, a Houston company and its owner pleaded guilty in 2025 after shipping documents were falsified to conceal where advanced Nvidia chips were going. The seller has to decide whether to hold an order while it verifies a delivery it may never see.
🟥 What changes when customers rent the chips?
A customer can train a model on servers abroad without taking delivery of a single processor. A shipping record can locate the server. Identifying the customer running a workload requires a different set of records.
U.S. Commerce Department guidance says some transactions involving cloud providers can trigger a license requirement when suppliers know of a qualifying military-intelligence or weapons-of-mass-destruction end use. A separate proposal would have required U.S. cloud infrastructure providers and their foreign resellers to verify foreign customers’ identities. Commerce withdrew it in December 2025.
Remote access is back in the legislative debate. The Information reported this week that the White House asked lawmakers to add a provision to the proposed AI Overwatch Act requiring Commerce to restrict remote access to certain chips. The practical question is how far a provider would have to look past the company paying for the service.
🟥 Would tighter U.S. rules settle who gets the chips?
In January, Washington moved to case-by-case license reviews for exports of Nvidia’s H200 and comparable chips to China, subject to security conditions. Reuters reported last week that roughly 600,000 H200s were waiting in Hong Kong as Beijing withheld approval for large-scale imports.
Washington is weighing how much access to permit. Beijing is weighing how much dependence on U.S. chips to accept.
🟥 What does Europe actually control?
The distinction between the location of the servers and the identity of the customer matters in Europe, too. Deutsche Telekom’s Industrial AI Cloud in Munich runs on about 10,000 Nvidia GPUs, with data and operations kept in Germany. Adding capacity still depends on access to Nvidia hardware.
U.S. licensing guidance already looks beyond the delivery address for certain advanced chips. An export can require a license if the buyer’s ultimate parent is based in China, even when delivery is elsewhere. Future rules on remote access could raise a separate question for European operators, depending on whom the rules cover. Which customers could they serve from the hardware they already have?
Signal
Watch who Washington asks to verify the end user. If distributors and cloud providers are expected to look past the buyer on the contract, the rules will have to say what counts as proof and when a company must hold a shipment or refuse access to computing capacity.
In the next edition of Altair News, we’ll look at how proposed U.S. restrictions on remote access could affect who can rent advanced AI chips.
Subscribe to Altair News for free.
Credit
AI-generated illustration / Altair Media US
Caption
An editorial illustration of Jensen Huang alongside an AI chip on a shipping manifest. “Know Your Customers” points to a distinction at the heart of U.S. export oversight. The company that buys the chip and the company that uses its computing power can be different businesses.
