While Washington draws red lines on chip exports, Beijing is drawing blueprints with the same chips on the other side of those lines.

The Summary

The Signal

The Malaysia data center story reveals something more interesting than simple sanctions evasion. Chinese tech giants are running compute infrastructure offshore not as a temporary workaround, but as architecture. When you build data centers in a third country to access chips you can't import directly, you're not breaking the rules. You're showing the rules don't govern the territory that matters.

The scale makes it real. $3.2 billion in active Chinese AI infrastructure projects are specced around the exact Nvidia chips U.S. export controls tried to wall off. These aren't sketches. They're construction budgets, procurement timelines, and power contracts. The projects assume Nvidia access as a given, which means someone solved the access problem before the architects started drawing.

"Chinese data center projects are still being designed around restricted Nvidia processors, even as Beijing pushes domestic alternatives."

Here's what the dual-track strategy looks like in practice:

  • Public: Beijing champions Huawei chips and domestic semiconductor self-sufficiency
  • Private: Project leads specify Nvidia H100s and A100s in their technical requirements
  • Actual: Malaysian data centers provide the compute, Chinese companies provide the workloads

The offshore model solves more than chip access. It solves the skilled labor problem (Malaysia has engineers who know Nvidia's stack), the power problem (cheaper electricity outside tier-one Chinese cities), and the diplomatic problem (technically compliant with the letter of export restrictions). When authorities investigate whether hardware was diverted elsewhere, they're asking the wrong question. The hardware didn't get diverted. The entire compute operation did.

This creates a strange mirror of how U.S. companies are thinking about China exposure. American firms are moving manufacturing and assembly offshore to de-risk their China dependency. Chinese AI companies are moving their compute infrastructure offshore to de-risk their America dependency. Both moves assume the other side's restrictions are permanent and plan accordingly.

The Implication

Export controls that target hardware but ignore compute-as-a-service don't control much. If Chinese AI labs can rent Nvidia cycles from Malaysian data centers the same way they'd rent them from AWS, the restriction becomes a tariff with extra steps. Watch for U.S. policy to shift from blocking chip sales to tracking chip usage, which is exponentially harder to enforce.

For anyone building AI infrastructure, this is the new normal. Compute access matters more than compute ownership. Geography matters less than network latency. The companies winning the agent economy won't be the ones with the most chips in their home country. They'll be the ones with the shortest path to wherever the chips actually are.

Sources

The Defiant | Decrypt