TL;DR · 30-second read
The Short Version
- California’s top law enforcement officer has ordered Nvidia, the company whose chips power most artificial intelligence, to hand over information in an investigation into chips being slipped into China.
- A separate newspaper report says Nvidia was asked to sell chips to companies with ties to China’s military.
- Nvidia’s own forecast for next quarter, about $130 billion in sales, already counts on zero data center chip sales in China.
- So the real question is not lost sales. It is whether anyone can track where these chips end up after they are sold.
California’s attorney general has issued a subpoena to Nvidia as part of an investigation into the diversion of chips to China, CNBC reported. Separately, The New York Times reported that Nvidia was asked to sell chips to companies tied to the Chinese military and to Megaspeed. A subpoena is a legally enforceable demand for documents or testimony. It is an investigative step, not a finding of wrongdoing.
The scrutiny arrives days after Nvidia reported third-quarter fiscal 2027 revenue of $115.8 billion, including $2.4 billion of Hopper data center shipments to China. The company guided to $130.0 billion, plus or minus 3%, for the fourth quarter without assuming any data center compute revenue from China. Nvidia is also operating under interim CEO Colette Kress, appointed November 11 after Jensen Huang resigned.
Executive Summary
A state attorney general has opened a line of inquiry into how Nvidia’s chips reach China. Until now, the rules governing where advanced AI chips may go have been enforced mainly by federal export authorities. A California subpoena means a state investigator can now demand records about the same flows. Nvidia is headquartered in Santa Clara, California.
The financial timing matters. In its November 17 results, Nvidia told investors its $130.0 billion fourth-quarter outlook assumes no data center compute revenue from China. The company’s forecast is therefore already insulated from China. What the probe examines is not a revenue line Nvidia is counting on. It is the chain of distributors, server makers, cloud operators and resellers through which chips sold elsewhere could end up in China.
For data center operators, cloud buyers and AI infrastructure financiers, the practical implication is more documentation about end users and end uses. Nvidia’s role is also deepening downstream. It now guarantees partner financing and shares revenue with neoclouds, so questions about who ultimately runs its chips reach further into the infrastructure stack than they would for a conventional chip vendor.
China Is Already Out of Nvidia’s Forecast
Nvidia’s 8-K filed November 17 is unusually direct. The company expects fourth-quarter revenue of $130.0 billion, plus or minus 3%, and states that it is “not assuming any Data Center compute revenue from China in its outlook.” Nvidia’s CFO commentary discloses that shipments of Data Center Hopper products to China in the third quarter were $2.4 billion. Hopper is the generation of AI accelerators that preceded Blackwell and the newly shipping Vera Rubin. The $2.4 billion is roughly 2% of the quarter’s $107.6 billion in data center revenue.
For anyone modelling Nvidia, a probe about chips reaching China does not touch the number the company has asked investors to judge it by. China data center compute is already set at zero going forward. That moves the exposure from the forecast to the channel. Diversion, meaning chips sold to one buyer that end up with another the rules do not permit, happens after the first sale, through distributors, systems integrators, cloud operators and resellers. A subpoena in a diversion probe asks for records about that downstream flow.
The costs of that kind of exposure show up less in revenue than in friction. They include legal work, compliance staffing, end-use certifications, audit rights in contracts, and possibly slower fulfillment for buyers in jurisdictions regulators consider higher risk. None of these costs are quantified anywhere, and the outcome of the investigation is unknown. Still, this is where the financial and operational effect of a diversion probe would land.
A State Investigator Joins Federal Export Rules
Export controls on advanced AI chips are federal rules, set and licensed by the US Commerce Department. Compliance programs across the industry are built around those federal licenses and lists of restricted parties. A state attorney general’s subpoena adds a second investigator with its own compulsory process. For companies in the AI hardware chain, questions about where equipment ends up can now come from more than one direction, under more than one legal framework.
The New York Times report shows what that channel has to screen for. It names requests to sell to companies tied to the Chinese military and to Megaspeed. Being asked to sell is not the same as selling, and the report’s framing makes no claim of a completed transaction. The fair reading is that demand from restricted or sensitive buyers exists and reaches the vendor. How that demand is identified and refused, at the vendor and at every reseller after it, is what an investigation like this tests.
The timing adds a governance dimension. Nvidia’s November 12 8-K announced Jensen Huang’s resignation and the appointment of Colette Kress as interim CEO, alongside her CFO role, with Harvey C. Jones as the board’s first chair. Nothing in Nvidia’s filings connects the leadership change to the investigation. It does mean the company’s response will be led by an interim team while the board searches for a permanent chief executive.
Nvidia’s Deeper Downstream Role Raises the Stakes of Knowing the End User
Nvidia is no longer only a component vendor that ships to server makers and clouds and moves on. Its CFO commentary describes arrangements with neoclouds, the specialist GPU cloud providers, to serve “AI startups, model builders, enterprises, research organizations and sovereign customers.” Under these arrangements Nvidia may participate in revenue share generated from the neoclouds’ third-party customers. The company has also entered guarantees on land, power, shell and hardware financing, under which it may have to pay third parties if partners default. Its cloud service and data center lease commitments rose to $102 billion from $74 billion a quarter earlier, and supply commitments to $390 billion from $279 billion.
Nothing in the filings links these arrangements to China or to the investigation. The point is structural. A vendor that shares revenue from its partners’ end customers and backstops their financing sits closer to the end user than a chipmaker selling through distributors. That proximity is a commercial strength. It also makes it harder to argue that knowing who ultimately runs the hardware is entirely someone else’s job.
For operators and buyers, the likely practical result is tighter know-your-customer expectations throughout the stack. Neoclouds, colocation providers and resellers that host or resell Nvidia capacity should expect more questions about tenants, locations and resale terms, from the vendor and possibly from regulators. Firms with clean, auditable customer records are better placed. Those that rely on opaque intermediaries carry more risk.
Background
Nvidia designs the accelerators and networking systems that dominate AI data centers. Its data center business generated $107.6 billion of the company’s $115.8 billion in revenue in the quarter ended October 25, 2026. Advanced AI chips are subject to US export controls that restrict sales to certain countries and end users, including China, and the industry’s compliance programs have been built mainly around those federal rules.
Nvidia’s business model has also been extending downstream. Beyond selling chips and rack-scale systems such as Vera Rubin NVL72, the company now has commitments with neoclouds, revenue-share arrangements and financing guarantees that help partners secure land, power and data center capacity. It is navigating this period under interim CEO Colette Kress following Jensen Huang’s resignation on November 11, 2026. Source: Nvidia Asked to Sell Chips to Companies Tied to Chinese Military and Megaspeed (The New York Times), reporting on requests made to Nvidia; and California attorney general issues Nvidia subpoena in probe of China chip diversion (CNBC), reporting on the state investigation. Primary sources: Nvidia Form 8-K filed November 17, 2026; NVIDIA Announces Financial Results for Third Quarter Fiscal 2027 (Exhibit 99.1); CFO Commentary on Third Quarter Fiscal 2027 Results (Exhibit 99.2); Nvidia Form 8-K filed November 12, 2026; NVIDIA Announces CEO Transition (Exhibit 99.1).Sources

