How Chip Tracking Changes the Neocloud Business
Neoclouds and data center operators are being thrusted into new gatekeeper roles. The AI stack worldwide relies heavily on NVIDIA chips and with that comes stringent export controls. While virtually all players are impacted by NVIDIA chip telemetry (remote tracking) and new export control expectations, business risk and operational burden fall most significantly on those two layers.
Chip smuggling and circumvention through diversionary tactics have reduced the effectiveness of US export controls. The Super Micro and Nvidia Employees case highlighted ways to bypass border checks in Taiwan, before authorities caught on. The individuals allegedly circumvented controls during 2024 - 2025 through forged end-user documents, shell companies and routing gear through complex shipping routes - through Indonesia, Japan, and Hong Kong. Not unlike money-laundering.
Telemetry: The Next-Gen Prevention Method
Recently, NVIDIA rolled out new tracking software that checks a chip’s diagnostics to figure out its physical location, starting with Blackwell chips. Instead of relying on humans verifying where the chips are and where the servers are located, the tracked chip itself reports back metrics such as power usage, temperature, and data transfer speeds, providing clues on whether the chip is being used in an authorized data center. This is happening in anticipation of the upcoming export controls expansion, RASA (Remote Access Security Act), that aims to close the “cloud loophole”.
The Burden on Neoclouds and Data Center Operators
Neocloud operators (providers of GPU-as-a-service) face major risk. Unlike hyperscalers like Amazon or Microsoft, they do not have massive alternative businesses to absorb financial shocks. Neither do they have the the means to breakaway by customizing their own AI chips. Tighter tracking would require operators to run constant digital checks on user identities, payment data, and remote connections.
Data center operators are also very exposed even though they do not control chip use, software or AI models. They are separate landlords, providing the building, power, and cooling. However, in the event of breach, the operator risks being blacklisted (no more buying American components, software, or services) or having its equipment seized. And since they are also landlords of servers for banks and other enterprises, an investigation could lead to delays and reputation damage, that could trigger an exodus of other tenants. To prevent this, the operator must constantly audit electricity usage for unusual spikes and stringently check tenant identities to avoid hosting an “unauthorized” AI cluster.
Cybersecurity
Commercial concerns. Many neo-clouds pitch themselves as “sovereign clouds”. This means sensitive data of clients and AI workloads will remain isolated within domestic borders, and protected from foreign laws. However, if neocloud services end up transmitting client side-channel data to NVIDIA’s centers, this removes the illusion of isolation and it would be difficult to promise full data and physical protection.
Government concerns. Some foreign governments and security experts see this as a major cybersecurity liability. In the event of a breach, it could allow adversaries to trigger mass lockouts. Countries like China also worry about mass surveillance - its cybersecurity regulator has already questioned NVIDIA over concerns that these might function as backdoors for the US to track users and locations.
Even though NVIDIA has confirmed that back doors do not exist, and kill-switches are not deployed, software license revocation or cryptographic attestation blocks could create similar effects.
The Transformation Needed
First, operators need to master verification. This requires a transition from being a tech provider into an organization capable of seamlessly verifying every physical and digital metric scrutinized by authorities.
Second, operators need to guarantee sovereignty. To do that, some operators are strategically diversifying into alternative sources of chips paired with open-source software, and deploying geofencing. They also use isolated and dedicated server modules for particularly sensitive clients.
