Bloomberg reported on November 6, 2026 that Nvidia now requires Southeast Asian customers who want Nvidia Cloud Partner (NCP) status to enroll in a program that tracks the locations of their AI chips. The reporter is Mackenzie Hawkins, and the story is titled Nvidia Boosts Chip Tracking in Southeast Asia to Fight Smuggling. It cites unnamed people familiar with the matter. Bloomberg's page is paywalled and we could not open it. We could not find an on-the-record Nvidia statement on this specific requirement either.
So this post follows the "what is reported versus what is confirmed" format. It states the claim, gives the confirmed background from Nvidia's own December 2025 post, and lists what readers can check.
TL;DR: reported vs confirmed
| Question | Status |
|---|---|
| Does Nvidia require Southeast Asian NCP applicants to join a tracking program? | Reported by Bloomberg via anonymous sources. Not confirmed by Nvidia in sources we could read. |
| Is the program called "fleet intelligence"? | Per secondary summaries of the Bloomberg report. We could not verify the name in a primary source. |
| Does Nvidia have a GPU fleet monitoring service? | Confirmed. Nvidia's December 10, 2025 blog post describes an opt-in, customer-installed service. |
| Does that service verify chip location? | The Nvidia post does not mention location verification. It describes grouping nodes into compute zones. Other outlets reported a latency-based location feature in 2025. |
| Does Nvidia say it can disable GPUs remotely? | No. The same post says its GPUs have no hardware tracking technology, kill switches or backdoors. |
| What do NCP partners get? | Priority access to Nvidia chips and a public stamp of approval, per the Bloomberg summaries. |
What exactly is reported?
According to summaries of the Bloomberg story, Nvidia wants applicants for NCP status in Southeast Asia to join a system the report calls fleet intelligence. The system tracks the location of the AI accelerators that a company owns, among other metrics. The NCP label is the incentive: it brings priority access to Nvidia's most wanted chips and a visible mark of approval.
The summaries also say the system does not automatically force users to share their location data with Nvidia or with the US government. That is a key detail and we could not check it against the original text. If it holds, the program is a condition of getting a partner label, not a legal export rule.
The reason is China. US export rules limit which advanced chips can go to Chinese buyers, and a series of smuggling cases through Southeast Asia has put pressure on Nvidia and on the region. Our earlier coverage tracked several of them, including a Malaysian seizure of 30 AI servers worth RM49.5 million and a guilty plea from a Super Micro contractor over AI server smuggling.
What Nvidia has confirmed on the record
On December 10, 2025, Nvidia published "Opt-In NVIDIA Software Enables Data Center Fleet Management." Facts from that post:
- It describes the service as "an opt-in, customer-installed service."
- Customers install a client agent that streams node-level GPU telemetry to a portal hosted on NVIDIA NGC.
- Telemetry covers power draw, utilization, memory bandwidth, interconnect health, thermal hotspots, airflow and errors that may signal failing parts.
- The telemetry is read-only. The software cannot change GPU configurations.
- The client agent is slated to be open-sourced for transparency and auditability, with no license named.
- Nvidia states that its GPUs do not have hardware tracking technology, kill switches or backdoors.
The post does not describe location verification. Tom's Hardware, however, reported in 2025 that the software can detect the physical location of processors, as a possible smuggling deterrent, though it was opt-in. Other outlets said location tracking was not included. The two accounts conflict, and Nvidia's own post is the better primary source. Nvidia also told Reuters in 2025 that no feature lets Nvidia or anyone else disable a GPU.
Secondary coverage says location estimates rely on timing: the delay between a GPU and Nvidia servers hints at distance. That is a known technique, but Nvidia has not published the method in the post we read.
Why would Nvidia do this now?
Pressure has built across several fronts in 2026:
- Smuggling cases. Customs seizures and court cases show chips reaching China through intermediaries. See our report on Tencent leasing 100,000 chips from Oracle, which shows the legal side of the same gap: renting remote compute instead of owning chips.
- Customer-mix pressure. On November 3, 2026 the Financial Times reported that Nvidia told cloud clients to sell at least 51 percent of services to non-Chinese customers. Reports disagree on what the 51 percent measures. Our post on that FT report separates claim from fact.
- Policy. Washington has debated remote-access rules for years. See our coverage of the August 2026 export control and remote access debate.
- Data points on stockpiles. Reuters reported a pile of H200 chips in Hong Kong that Beijing has not approved for import.
Reporting on the Bloomberg story also mentions a probe involving Megaspeed, a Southeast Asian data-center company. See our post on the Megaspeed spy-plot report for what is claimed and what is not. We cannot say whether the new requirement is tied to any single company.
Timeline: how the story reached this point
The dates below come from the sources linked in this post. We list only events we could tie to a named source.
| Date | Event | Source |
|---|---|---|
| December 10, 2025 | Nvidia publishes its opt-in fleet management post and says its GPUs have no kill switch or hardware tracking | Nvidia blog |
| December 2025 | Outlets report that the monitoring software can estimate chip location, a claim other outlets dispute | Tom's Hardware and others |
| November 3, 2026 | FT reports Nvidia told cloud clients to sell at least 51 percent outside China | FT, via Forbes and our earlier post |
| November 6, 2026 | Bloomberg reports the Southeast Asia NCP tracking requirement | Bloomberg, anonymous sources |
Read the table as a direction of travel, not as proof of one policy. Each step is a separate claim from a separate source. The newest one has the weakest sourcing, because it relies on unnamed people and no company statement.
How should you weigh a one-source report like this?
A report from a major outlet with a named reporter is a credible signal. It is still a single chain of anonymous sourcing. Three habits help.
First, separate the mechanism from the motive. The motive, stopping smuggling, is well documented through court cases and customs seizures. The mechanism, a partner-status condition, is the part that needs confirmation.
Second, watch for wording that hedges. Phrases such as "among other metrics" and "does not automatically force users to share" suggest the program is broader than location alone and softer than a mandate. Those are exactly the details that can change between a leak and a written policy.
Third, wait for the second source. A partner announcement, a regional regulator note or an Nvidia reply to a press question would settle most of the open points. Until then, treat this as a credible report, not a settled rule.
What this means if you build on rented GPUs
Most readers do not buy GPUs. They rent them from neoclouds. Three practical points:
- Supply. NCP status brings priority access to scarce chips. If monitoring is the price, Southeast Asian providers will likely accept it. That could make their supply more stable, or more tightly controlled.
- Telemetry. Nvidia describes the service as read-only GPU health data. Ask any provider you use what is sent off its site, how the agent is licensed and whether the code is open for audit.
- Compliance. If your workloads include customers or models linked to China, check your provider's terms. Partner status is not a legal safe harbor, and the reported program, if real, is a commercial condition, not law.
Do not assume more than the evidence shows. We found no evidence in the sources we could read that Nvidia can switch chips off. Nvidia says it cannot.
How to check the claim yourself
- Read Bloomberg's report if you have access and note which statements are attributed to sources versus Nvidia.
- Re-read Nvidia's fleet management post and look for any update that mentions location or partner requirements.
- Look for an Nvidia statement to Reuters, Bloomberg or its newsroom on the Southeast Asia requirement.
- Watch for a release of the open-source client agent. Its code would show exactly what data is collected.
- Ask your GPU cloud provider whether it is an NCP and whether it uses the fleet service.
What is still open
- Whether the requirement applies only to Southeast Asia or will spread to other regions.
- Whether location data goes only to Nvidia or may also reach US authorities. The reported answer is that it is not automatic.
- How accurate latency-based location can be, and how a determined smuggler could evade it. Nvidia has not published accuracy numbers.
- Whether existing partners must enroll or only new applicants.
For security teams, an agent that monitors hardware is also a new piece of software on your network. The AgentBeam platform from the explainx.ai team is one way to check what AI agents and tools can do before they act.
Related reading
- Nvidia tells cloud clients to sell 51% to non-Chinese customers: FT claimed vs verified
- Tencent leases 100,000 AI chips from Oracle
- Malaysia seizes 30 AI servers in a chip smuggling probe
- Super Micro contractor guilty plea over Nvidia server smuggling
- Export controls and remote access, August 2026
- Nvidia H200 stockpile in Hong Kong
Details are accurate as of November 6, 2026. This story rests on anonymous sourcing and may change when Nvidia comments.