WSJ: China Is Quietly Bypassing U.S. AI Chip Controls Through a Legal Three-Step Loophole Network

A new Wall Street Journal investigation reveals that Chinese companies banned from purchasing advanced U.S. AI chips are still accessing Nvidia’s most powerful processors — legally — by exploiting gaps in U.S. export controls.

The system is not a single workaround; it’s a chain of loopholes that, when combined, allow blacklisted Chinese firms to train and run advanced AI models using Nvidia’s latest Blackwell chips without ever importing the chips into China.

The investigation exposes a core weakness in the current American approach: the U.S. restricts where chips can be sold — but not who can ultimately use them.

Below is what is happening, step by step.

1. Chinese Firms Can Buy U.S. AI Chips — As Long as They Buy Them Inside the United States

The WSJ found that Nvidia has been shipping its export-controlled Blackwell GPUs to a U.S.-registered company called Aivres Systems.

Aivres is legally American — but it’s 100% owned by Inspur Electronic, a China-based company on the U.S. Entity List (i.e., officially blacklisted).

Under current law:

• Banned Chinese companies cannot buy advanced AI chips in China.

• They can buy them inside the United States.

This is one of the biggest blind spots in the existing export-control framework, and it is fully legal.

So Aivres buys Blackwell chips domestically — no export license needed.

2. The Chips Are Then Exported to Indonesia — Legally

After Aivres acquires the chips inside the U.S., they are sold to Indosat, a cloud-services provider in Indonesia.

Why Indonesia?

Because U.S. export controls don’t apply to most third-country destinations unless the company is majority-owned by a Chinese entity. Indosat is majority-owned by a joint venture between:

• a Qatari firm, and

• a Chinese firm

— but the JV is not considered China-based, so the sale to Indosat is legal.

At this point, the chips have left U.S. soil but are still in compliant channels.

3. A Blacklisted Chinese Firm Rents Access to the Chips Abroad — Also Legal

This is the most important loophole.

Through Aivres, Indosat identifies a Chinese client:

INF Tech — a company banned from purchasing AI chips.

INF Tech cannot legally import or buy them.

But under current U.S. rules, it can:

• rent compute

• lease GPU clusters

• or run AI models hosted outside China.

So INF Tech rents access to the Indosat cluster — the same cluster using Nvidia Blackwell chips sourced from Aivres.

INF Tech now gets full access to the compute needed to train advanced AI systems.

Everything so far is legal.

No sanctions breached. No export license avoided. No illicit transfer.

Just loopholes — stacked cleverly.

Why This Matters: The U.S. Restricts Chips, Not Compute

The U.S. has framed export controls around hardware flows, not compute access.

This makes sense for a world in which:

• chips are physical

• access must be on-site

• compute is bounded by borders

But in AI cloud environments:

• the user and the chip do not have to be in the same country

• blacklisted firms can train models remotely

• U.S. rules restrict shipments but not cloud access

This is how China’s banned firms are still developing advanced AI — by using U.S. GPUs located outside China.

What Experts Say Needs to Change

Policy analysts and former Commerce officials argue three fixes are needed immediately if Washington wants to prevent compute leakage:

1. Close the U.S.-based shell-company loophole

Ban Chinese-owned firms operating inside the U.S. from purchasing restricted chips.This can be done by expanding ICTS (Information and Communications Technology and Services) regulations.

2. Tighten export controls to Southeast Asia

Require U.S. export licenses for large AI chip shipments to “untrusted” destinations — especially countries where China has corporate or political influence.

3. Close the remote-access loophole

Ban China-based entities from using advanced AI chips — not just from buying them.

This would extend export controls to:

• cloud compute

• remote inference

• remote training

• cross-border GPU rental

— the areas where current policy is weakest.

Without this, China can continue building advanced AI capabilities while staying technically compliant.

The Bottom Line

This investigation shows that the U.S. has built an export-control regime designed for the 20th century:

• stopping shipments,

• tracking physical goods,

• controlling geography —

while ignoring how AI actually works today:

• dispersed compute,

• multinational cloud providers,

• remote access,

• joint-venture ownership webs.

As long as these loopholes remain open, China can continue developing frontier AI models using U.S. chips — without ever importing them.