A GPU can stay in one country while a customer in another pays to use it. In November 2026, Nvidia reportedly required Southeast Asian applicants for its Cloud Partner program to enroll in chip-location tracking. The machine’s address is the easier part.

Key takeaways

  • Reuters reported in December 2025 that Nvidia privately demonstrated unreleased technology able to indicate the country where its chips were operating.
  • Bloomberg reported on November 6, 2026, that Southeast Asian applicants to Nvidia’s Cloud Partner program had to enroll in AI-chip location tracking.
  • Sources said Nvidia made two introductions between Alibaba and Alice Huang in 2022.
  • Sources said Nvidia verbally told cloud-compute clients in November to sell more than 51% of their services to non-Chinese companies.
  • Nebius reported $980.8 million in third-quarter revenue, a 571% year-over-year increase.

Nvidia’s reported partner requirement asks GPU-cloud operators to document where their chips run, while its scrutiny of customers and intermediaries extends beyond the hardware sale. A shipping record identifies a buyer and a destination; a running cloud service also has a host, a scheduler, a reseller and an effective user. Those can be different parties.

The introduction outlasted the sale

In 2016, Alibaba’s AliCloud partnered with Nvidia to develop a GPU-based computing platform in China. The commercial question was how to put processors into a cloud that could rent their capacity. Nvidia could sell hardware and support the platform without treating every subsequent workload as part of the original transaction.

Scarce capacity made introductions more consequential. Sources reported that Nvidia made two introductions connecting Alibaba with Alice Huang in 2022, helping Megaspeed secure Alibaba as a customer. An introduction is not evidence of prohibited end use. It does, however, leave a record of who helped bring a customer to a provider—information that matters differently when authorities later ask who controlled the relationship.

A subsequent investigation connected Huang to gaming firms and hundreds of shell companies in a reported chip-trading network; its links to Prince Group figures were described as alleged. Megaspeed said Huang left in August 2025, has no current role, and that the company runs a legitimate and lawful business while it conducts an internal probe. About 55 tech-news articles in the record covered Huang from 2024 to 2026. That measures scrutiny, not guilt. Neither a customer introduction nor the volume of reporting establishes who ultimately used a chip.

Location verification answers only one question

Reuters reported in December 2025 that Nvidia had privately demonstrated unreleased technology capable of indicating the country in which its chips were operating. Bloomberg reported on November 6, 2026, that Nvidia was requiring Southeast Asian applicants for Nvidia Cloud Partner status to enroll in a program tracking their AI chips’ locations. Reuters described a capability; Bloomberg described a sales-channel qualification. Neither established whether Nvidia uses the demonstrated technology for enrollment or how extensively the program collects location data.

Nor does the reported enrollment requirement amount to a universal government telemetry mandate or a formal export-license condition. U.S. officials had previously been reported to be considering disclosures of large cloud-compute purchases and permission requirements for certain services supplied to Chinese customers. Those reports described proposals and plans, not proof that every cloud provider acquired a new obligation. Nvidia also said in December 2025 that it had not seen substantiation or received tips about chip smuggling through data centers outside China.

Even with a chip’s country confirmed, a provider must know which account received capacity, who introduced or resold that account, and whether another party effectively directs the workload. A cloud provider schedules jobs across machines and networks under customer accounts; a chip cannot identify the person at the other end of a rented service. Building a compute chain of custody therefore means preserving evidence across those handoffs, with limits on what each piece of evidence proves.

Nvidia can ration the channel before a regulator rules

Sources reported in July that Nvidia had cut its authorized Asian AI-chip clients by more than half and intensified diligence in Singapore, Malaysia and Japan to prevent diversions to China. In November, sources said Nvidia had verbally told cloud-compute clients to sell more than 51% of their services to non-Chinese companies. A reported verbal instruction is not a published universal policy. It nevertheless asks a cloud to account for its customer mix, not merely the destination of its latest shipment.

Nvidia was also reported to offer some young providers a backstop by renting unused GPUs in exchange for a share of revenue. That arrangement gives the chip supplier an interest in whether cloud capacity finds customers after delivery. Nvidia can support utilization while selecting which operators receive inventory and commercial backing. Export-control scrutiny and capacity allocation meet at the partner application, before any dispute over a particular workload has been resolved.

A customer record now sits beside the financing record

Nebius reported $980.8 million in third-quarter revenue, up 571% year over year, while raising its planned 2026 capital expenditure. Sources also said CoreWeave, Nebius and Nscale had effectively been shut out of the bond market as lenders and investors demanded tougher terms. Those sources did not identify chip-location conditions among the terms. A cloud operator cannot treat access to supplier-supported capacity as an incidental compliance matter: it is committing capital while its route to inventory and customers remains subject to another company’s decisions.

Frequently asked questions

Which Southeast Asian cloud operators were required to enroll in Nvidia’s tracking program?

The reporting cited in the piece does not name the applicants or say how many operators were affected. It describes a qualification requirement for Southeast Asian applicants to Nvidia Cloud Partner status, not a published list of enrolled companies.

When did Nvidia’s reported Cloud Partner tracking requirement take effect?

The piece dates Bloomberg’s report to November 6, 2026, but provides no effective date, rollout schedule, or deadline for applicants. It also does not say whether existing partners were included.

How is the reported requirement that more than 51% of services go to non-Chinese companies measured?

The sources described a verbal instruction, but the piece does not specify whether the threshold refers to revenue, compute hours, contracts, customers, or another measure. Nor does it describe an audit or enforcement method.

Did lenders identify Nvidia tracking or customer-control conditions as a reason for tougher financing terms for GPU clouds?

No. The sources said CoreWeave, Nebius, and Nscale faced tougher bond-market terms, but did not identify chip-location conditions among those terms.

Reported milestones in Nvidia’s cloud-control shift

  • 2016 — Alibaba’s AliCloud partnered with Nvidia to develop a GPU-based computing platform in China.
  • 2022 — Sources reported that Nvidia made two introductions connecting Alibaba with Alice Huang.
  • August 2025 — Megaspeed said Alice Huang left the company and had no current role.
  • December 2025 — Reuters reported that Nvidia privately demonstrated technology capable of indicating the country where its chips were operating.
  • November 6, 2026 — Bloomberg reported that Southeast Asian applicants for Nvidia Cloud Partner status were required to enroll in AI-chip location tracking.

A cloud can document the country of a rack and still leave the decisive relationship undocumented. Nvidia’s reported partner requirement puts an address on the machine; the customer introduction, account record and capacity contract can show who holds the keys.