Nvidia Now Demands GPU Tracking From Cloud Partners Hunting Smuggled Chips

Cloud providers in Southeast Asia must now enroll in Nvidia's Fleet Intelligence tracking system if they want official Nvidia Cloud Partner status, Bloomberg reports — a direct response to restricted AI accelerators leaking into China through regional smuggling networks.

The NCP designation carries real weight. It brings Nvidia's public endorsement and preferred access to its scarce AI chips, which hands the company leverage: want the chips, join the tracking program. People familiar with the new requirements confirmed the enrollment mandate. Nvidia was not immediately available for comment.

What does Fleet Intelligence actually see?

Nvidia introduced Fleet Management last December, primarily as a data center management tool. The software gathers information about Nvidia hardware deployed across participating companies' infrastructure — including the locations of individual accelerators. That gives Nvidia a live picture of where its chips sit and run after they leave the supply chain.

But it's not a universal surveillance dragnet. Participation has stayed voluntary outside the NCP requirement, and Nvidia doesn't automatically receive location data from every customer running the software. The company also hasn't clarified two things: whether existing Cloud Partners must enroll to keep their designation, and whether the requirement will eventually extend beyond Southeast Asia.

The tracking capability itself isn't new. Tom's Hardware reported last December that Nvidia had built software-based GPU tracking using GPU telemetry and network latency measurements to estimate where processors are operating. Nvidia framed it then as a customer-installed agent for monitoring hardware health, integrity and inventory.

Why the sudden enforcement?

The U.S. bars Chinese companies from buying advanced AI accelerators without a license. Authorities nonetheless believe billions of dollars worth of restricted hardware has reached Chinese buyers through third-party networks, with Southeast Asia serving as a key transit point.

The evidence is specific. A C4ADS investigation — examining transactions between 2022 and 2025 — identified several routes used to move restricted processors into China: purchases through research initiatives, shell companies, and shipments routed through other Southeast Asian countries. The nonprofit traced $13.4 million worth of Nvidia A100, H100 and AD102-series GPUs moving through regional trade routes. Its report also linked $4.6 billion in hardware purchases to Singapore-based Megaspeed International.

Megaspeed has become Nvidia's most awkward case. The company has been under U.S. and Singaporean investigation for more than a year over suspected chip diversion, and it denies any wrongdoing. Last week, Nvidia CEO Jensen Huang defended his dealings with the firm, describing them as "completely normal," and adding: "we've studied all their deployments of our GPUs and counted every one."

The pressure is showing elsewhere too. In April, Megaspeed's Malaysian data centre landlord terminated its contract and reallocated 68.4 megawatts of capacity to another cloud provider.

Who else has been caught in the net?

Megaspeed is hardly alone in illustrating how hard AI hardware is to police. Earlier this year, U.S. prosecutors accused people associated with Super Micro of helping divert roughly $2.5 billion worth of Nvidia-powered servers to China through Southeast Asian intermediaries. The alleged scheme used falsified paperwork and dummy server shells, with genuine hardware shipped onward after inspections. Super Micro has since faced additional scrutiny over export compliance following those inspections.

Nvidia has already tightened its own gates. The company has reportedly removed half of the Asian companies on its pre-approved buyer list and told cloud providers to keep at least 51% of their computing services with customers outside China.

For cloud providers across the region, the calculus is now blunt: transparency about where every accelerator sits, or no official partner status and no preferred chip access. Whether Nvidia extends the mandate worldwide — and whether it proves enough to satisfy Washington — will shape the next phase of export enforcement.