In the past, the United States restricted exports of advanced GPUs by focusing on the “physical chips”—finding out who they were sold to and where they were shipped.

But today’s AI business has long since broken through those physical boundaries.

GPUs can be openly housed in data centers in Singapore, Malaysia, or the Middle East, while customers in China can simply rent that computing power remotely through cloud platforms to train large models.

The chips never crossed a border, but the computing power did.

This raises a question that strikes at the heart of regulation: What exactly is the United States restricting—the chips themselves, or the “computing power” behind them?

This is no theoretical scenario. The United States has already begun imposing restrictions on “remote access” through the terms of certain export licenses, and Congress is actively pushing to expand regulatory authority further.

If the regulatory net tightens, those affected will be far more than GPU traders. Data centers in Southeast Asia, GPU rental platforms, international cloud service providers, and even AI model training companies will all face significant compliance risks.

The business logic has changed completely.
In the past, the most important things in the GPU business were supply, price, and logistics.
In the future, doing business in computing power will also mean finding out: Who is the customer? What is the end use? Who actually controls the computing power?
When selling chips, the question used to be “Where is the product going?” In the future, selling computing power may also mean asking “Who is using it?”