Posted on — Leave a comment

Know your tenant: US chip rules turn Southeast Asia’s data centres into compliance desks

For most of this decade, Southeast Asia’s data centre operators asked prospective tenants a short list of questions: how many megawatts, how long a lease, how soon. Since November 16, that list has grown, and the question that matters most is no longer how much power a customer needs but who ultimately owns it.

That is the clearest reading of a new note from BMI, the research unit of Fitch Solutions, on the US Commerce Department’s latest ruling on advanced AI chips. The rule requires licences for shipments of specified advanced semiconductors to Indonesia, Malaysia, Singapore, Thailand and Brazil. More pointedly, it stretches export controls beyond the chip itself to whoever rents the computing power it produces.

Also Read: The Nvidia clampdown is a warning for Southeast Asia’s AI boom

The ruling arrives in the wake of the Megaspeed affair, which put a Singapore-based Nvidia cloud partner under investigation over suspected chip diversion to China (the company denies wrongdoing) and has already cost Jensen Huang his job at Nvidia. Washington’s message to the region is blunt: prove the leak does not run through your server halls.

A 750-chip tripwire, measured in compute

The sharpest clause targets data centres outside the US that supply more than 12 million aggregate total processing performance (TPP), a yardstick of raw computing muscle, to companies headquartered in mainland China, Macau or other restricted destinations. Cross that line and the operator needs a US licence.

BMI puts the threshold at roughly 750 Nvidia H100-class chips, with capacity aggregated across linked sites. That last detail closes the obvious escape hatch: splitting a cluster across several halls to stay under the line will not work.

The goal is to shut the loophole that lets companies in restricted countries rent advanced American computing overseas without importing a single chip. Rules on direct chip sales to China are unchanged; what changes is oversight of the ultimate user. The ruling was slated to take effect on November 19, upon publication in the Federal Register.

Nvidia and AMD become the region’s KYC desk

The immediate burden falls not on governments or even landlords, but on vendors. BMI says chip and server suppliers, including Nvidia and AMD, face the most immediate obligations. They need licences to ship specified chips to the five countries unless a deal qualifies for an exemption covering US-headquartered customers, which comes with monthly reporting.

Suppliers must also extract information from operators on the identity and ownership of their tenants, including anyone tapping compute through the facility. Operators that refuse will be treated as needing a licence until the concerns are resolved. In effect, the world’s dominant AI chip vendor becomes Southeast Asia’s know-your-customer (KYC) officer.

The net reaches upstream too. Foreign foundries and packaging firms must treat advanced chips designed by companies in the five countries, or in other restricted jurisdictions, as controlled unless the designers hold the necessary approval. For a region that has built its semiconductor story on assembly, packaging and a fledgling design scene, that is fresh paperwork on every order.

Also Read: AI’s bottleneck economy is taking shape across Southeast Asia’s chip supply chain

The US Bureau of Industry and Security (BIS) expects about 100 additional licence applications a year, each judged case by case. A hundred files is modest for a region pouring billions into server halls. Washington appears to be aiming a rifle rather than a shotgun, but nobody yet knows how often it will pull the trigger.

Two data centre markets are taking shape

BMI expects projects built around Chinese tenants to take the heaviest hit, while those serving US tech firms shelter behind the exemption. Existing operators supplying above-threshold compute to Chinese-headquartered customers get 90 days from the effective date to secure licences, which puts the deadline at February 17, 2027.

This is the split the region has been edging towards for months. Johor’s data-centre future is increasingly being written in Washington, Kenanga Research’s reading of the rules tilts Malaysia’s growth towards American hyperscalers, and Indonesia is paying the compliance bill for Asia’s AI chip race.

BMI’s more interesting call is that affected customers will be broken up rather than locked out. It expects them to drift towards smaller, fragmented and heavily documented compute arrangements. Translation: less capacity per contract, more lawyers per megawatt.

One carve-out deserves scrutiny. Compute used solely for commercial recommender systems, the algorithms that decide what you watch or buy next, is excluded as long as operators meet reporting requirements. Capacity inside the US falls outside this licensing clause altogether, a reminder that the rule polices the periphery, not the core.

Regulatory credibility is the new tax break

Licence decisions will also weigh how well the five countries enforce their own export controls, nudging governments to tighten domestic frameworks. Singapore has already signalled it will explore trusted access arrangements for its companies.

That turns regulatory credibility into a competitive asset, much as tax holidays and cheap land once were. A country that convinces Washington its corridors are clean earns faster licences and calmer investors. Given the region’s chip-smuggling problem, that credibility will not come cheap.

Don’t bank on a Shenzhen reprieve

BMI expects the substance of the rule to survive a public consultation that closes in January 2027, though details could change. Pressure in Washington is running the other way: three chip-control measures are reportedly included in the Senate version of the fiscal 2027 defence bill.

Also Read: Why Southeast Asia cannot build sovereign AI on borrowed choices

The research house did flag two possible softeners: President Donald Trump’s meeting with Chinese President Xi Jinping in Shenzhen on November 19, and the January 10, 2027 expiry of the current US-China trade truce. Either could tempt Washington to trade concessions, and the administration has bent before, permitting case-by-case licensing for Nvidia’s H200-class chips in January 2026.

Still, BMI warns against assuming a repeat. Its advice for the first quarter of 2027 is to plan for full compliance by February 17 and treat any relief as upside, not baseline.

For founders, operators and investors weighing a stake in the region’s data centres, that is the real takeaway. Southeast Asia’s compute pitch used to rest on land, power and proximity. Ownership structures and the identity of whoever sits behind each rack now belong on that list, and unlike power shortages, they cannot be fixed with a bigger substation.

The post Know your tenant: US chip rules turn Southeast Asia’s data centres into compliance desks appeared first on e27.

Leave a Reply

Your email address will not be published. Required fields are marked *