Key Takeaways

  • Nvidia changed its accounting to report revenue by customer headquarters rather than physical shipping drop points, obscuring chip deliveries to logistical pass-through hubs like Singapore and Malaysia.
  • Direct China revenue fell to roughly 8% of Nvidia's total sales, consisting primarily of automotive contracts and gaming hardware rather than high-end data center silicon.
  • United States buyers account for more than 60% of Nvidia's revenue, while Taiwanese buyers make up the low 20% range.
  • Rumors that illegal Chinese smuggling accounts for 20% of Nvidia's top line are mathematically implausible, but Rasgon argues that any actual leak occurs through Taiwanese intermediaries, not American tech giants.

Headquarters Accounting Masks Physical Hardware Flows

Tracking where artificial intelligence silicon goes has become an obsession in Washington and Wall Street. But reading Nvidia's earnings filings will not show you where the physical servers sit. Bernstein analyst Stacy Rasgon points out that the chipmaker changed how it accounts for regional numbers: “The company does report revenues by geography, and they do this now, they made a switch a little while ago to report by the location of the headquarters of the companies.” (play from 12:14)

Under this system, Nvidia groups its geographic numbers into the US, Taiwan, China (including Hong Kong), and an "other" category. When a multinational tech giant buys chips and drops them into data centers across Singapore or Malaysia, that revenue gets recorded under the buyer's home base. Rasgon notes that this reporting choice wipes out visibility into regional hubs: "So on that basis, like Malaysian and Singapore basically any of the they report now US, Taiwan, China, including Hong Kong, and and other. And other would be Singapore and Malaysia and all these other ones."

If you want to track where server racks actually dock, the 10-Q cannot help you. It records corporate billing addresses, not shipping manifests.

The Taiwan Intermediary Pipeline

Washington hawks frequently float claims that vast quantities of banned GPUs cross into mainland China through illicit black markets, with some critics suggesting illicit sales hit 20% of total revenue. Rasgon walks through the customer distribution to test that thesis. US buyers drive over 60% of top-line revenue, and direct Chinese sales sit near 8%. As Rasgon explains, “The direct sales to Chinese customers, so China, including Hong Kong, I think last quarter was roughly 8% of their sales. And that will include a big chunk of their gaming and auto revenues as well.” (play from 14:14)

That leaves Taiwanese buyers, who represent the low 20s percentage of sales. If hardware is leaking across borders at any meaningful scale, the math points to one specific channel. “My assumption would be that if there's like stuff that is actually getting smuggled into China, it's probably not from the US customers, right?” (play from 15:07) Rasgon says. “It's probably not from the Chinese customers, you're not selling stuff to them anyways, or not very much. And the Malaysian and Singaporean customers is very small, so it's got to be the Taiwanese customers.” (play from 15:16)

Leaking thousands of water-cooled server racks requires industrial logistics. The idea that rogue actors smuggle billions of dollars of data center hardware without detection stretches credulity, but Taiwan remains the sole corporate channel big enough to host leaks.

Cloud Access Outweighs Physical Smuggling

While regulators fixate on physical shipping containers, Chinese entities bypass export restrictions through software. Chinese developers can legally rent high-end compute hosted outside their borders. Rasgon highlights this regulatory loophole: “So right now it is legal for a company in China to access, you know, cloud services from out outside of China, and they want to put a stop to that. I don't know how big that is, right? I don't know.” (play from 15:09)

Plugging a leak in physical silicon creates headlines, but closing international network endpoints across foreign clouds is nearly impossible to police. Regulators chasing customs manifests are fighting the previous war while the real workload runs over remote APIs.

What to Do With This

Audit your cloud provider and compute procurement agreements this week. Check whether your contracts contain geographical routing restrictions or secondary-use clauses that bar non-US developers from leasing cluster capacity. If you resell compute or host foreign API traffic, write an explicit compliance policy separating physical hardware locations from overseas API access before export regulators draft new offshore cloud rules.