Malaysia to Force Data Centers to Disclose GPU Imports and Power Use

Malaysia will write into law a requirement that data center operators disclose their electricity consumption and the number of GPU chips they import, Investment, Trade and Industry Minister Johari Abdul Ghani announced on Nov. 12. The measure is aimed squarely at smugglers who rent space in Malaysian facilities, import restricted Nvidia-class hardware, and quietly re-export it.

Johari, speaking in remarks carried by the state news agency Bernama, said the government also wants to know who operates each data center, which hyperscalers rent its capacity, and where every facility sits. Of Malaysia's 117 approved data center projects, 55 are now operating, he said, and monitoring must continue after approval rather than end at the ribbon-cutting.

What does the disclosure law actually change?

The core idea is arithmetic. Importers bring GPUs into Malaysia today without any permit, so the state has no registry of what hardware entered the country or where it landed. Mandatory disclosure of GPU counts and electricity use would, in principle, let regulators spot a facility whose power draw does not match its declared hardware. Tom's Hardware, which noted the approach, also flagged its limits: real-world chip power consumption varies, and without physical audits the numbers rest on self-reporting.

The Digital Ministry is separately reviewing whether existing laws can be used against re-exports of GPUs restricted by the United States. Johari was direct about the motive: keeping Washington's trust. "When the US imposes a restriction, as a country, we want to make sure that whatever chips arrive here are meant for our own use," he said. He added that the government could confirm to US authorities which data center is using the chips in question.

The minister gave no timeline for the legislation and did not say which agency would enforce it. That gap matters. A disclosure law without an enforcement body is a suggestion with better paperwork.

The smuggling backdrop

Since July 2025, exporting, re-exporting or transshipping high-performance US AI chips through Malaysia has required a trade ministry permit — but imports still do not. That asymmetry leaves a window: a chip can arrive legally, sit in a rented rack, and leave the country before anyone counts it.

On Nov. 13, the trade ministry said it had never cleared Megaspeed, the Nvidia cloud partner at the center of the Jensen Huang scandal, or its Malaysian affiliate Speedmatrix. It is working with customs to determine whether chips Speedmatrix brought into the country were subsequently re-exported. Megaspeed denies any wrongdoing.

Paying for power you reserve, not power you use

Johari also repeated a separate grid rule that will pinch operators' economics. Companies will be billed for the grid capacity they book, not the capacity they draw. A site that reserves 100 megawatts but consumes 50 will still pay for the idle 50.

The rule targets a habit common in the industry: hyperscalers reserve far more capacity than they need, sometimes years ahead, and leave it unused for one to three years. For a country whose 117 approved projects represent a serious bet on AI infrastructure, stranded grid reservations are a real cost — capacity locked away from other industrial users while it waits for racks that may never arrive.

Why Malaysia is moving now

Malaysia has become a convenient waypoint in the flow of restricted AI hardware through Southeast Asia, and Washington's export controls have made transit jurisdictions a enforcement priority. Johari's pitch is essentially commercial: a country that can certify where every restricted chip sits is a country hyperscalers can keep building in. One that cannot may find its approvals subject to rather more American attention.

The coming months will show whether the legislation arrives with an enforcement mechanism attached, and whether the Megaspeed customs review turns up re-exports. For now, Malaysia's 62 data center projects still in the pipeline face a future of counted GPUs and billed megawatts.