BIS Remote-Compute Rule Takes Effect on Summit Day: A 12-Million-TPP Line, Five Countries Licensed, and Southeast Asia's Data Centers on the Clock

Text
iHOLYSHITVISUAL / TECHIndustry changesiHOLYSHIT / ILLUSTRATION
iHolyShit illustration

The trilateral declaration never mentioned technology; what took effect on summit day was BIS's remote-access rule — a 12-million-TPP licensing line, five new licensed countries and a foundry presumption. The real negotiating table is the February 17 license deadline.

On November 19 in Shenzhen, three heads of state shook hands and issued a peace declaration that does not mention technology once. That same day, a new rule from the US Commerce Department's Bureau of Industry and Security — "Addressing Remote Access to Advanced Semiconductors" — took effect. Its red line is precise: an overseas data center that provides more than 12 million in aggregate TPP of computing power — roughly 750 H100s or 330 B200s — to companies headquartered in China, other Country Group D:5 destinations or Macau now needs a license for the exports, re-exports and in-country transfers of the advanced chips and servers involved. The rule was released on November 16 and landed, pointedly, on summit day.

This is the core fact of the APEC technology storyline this newsroom is tracking: the summit texts contain zero technology clauses, and the real text was published in the Federal Register. The rule was written by the Treasury Department and the National Security Council, bypassing the Commerce Department — Treasury Secretary Bessent had publicly criticized Commerce for moving too slowly, and Commerce Secretary Lutnick was reported to have delayed new controls. The trigger was the Megaspeed pattern in Singapore: chips never leave the country, but the compute does. Nvidia moved first: on its November 17 earnings call, interim CEO Colette Kress announced the company had stopped shipping licensed H200 and other products to China, and that its $130 billion fourth-quarter guide assumes zero China data-center revenue. China's Ministry of Commerce responded on the evening of November 17 with strong dissatisfaction and firm opposition, vowing countermeasures; the Foreign Ministry called it economic coercion and long-arm jurisdiction.

The rule blocks three paths. Path one, remote compute: the new Section 744.24 turns on a "knowledge" standard — a data center that cannot produce customer due-diligence records is presumed to know — with no de minimis escape; data centers already supplying covered compute must obtain a license by February 17, 2027 — obtain, not merely apply. Path two, diversion: Indonesia, Malaysia, Singapore, Thailand and Brazil now need licenses to import advanced semiconductors; per the China Academy of Information and Communications Technology, Southeast Asia's key compute regions already exceed 4.4 gigawatts of capacity, with Singapore alone above 1.4 gigawatts — all of it inside the perimeter. Path three, fabrication: advanced chips are presumed controlled as they leave the fab, and design companies headquartered in controlled destinations can no longer rebut the presumption with transistor counts; they must get onto BIS's list of approved IC designers. On the ledger, Jefferies estimates the rule could cut $44 billion to $62 billion from Nvidia's data-center revenue next year — an analyst estimate, not company guidance, and Kress has publicly pushed back, saying there is no top-line impact. One gap remains open: cloud services from data centers on US soil sit outside this rule, and the Remote Access Security Act meant to close it is still working through Congress.

Layer one: Southeast Asia's data centers are nervous first

Layer one: Southeast Asia's data centers are nervous first. Two Singapore ministries have said they are studying the rule and seeking "trusted access" — the model of eating compute rents from a neutral position has been repriced by a license. Layer two: for foundries, the rational move is to refuse orders. Under a presumption regime, when compliance costs exceed order margins, no fab will carry a customer's risk for them; on November 19 Chinese chip-design stocks fell hard while foundry stocks rallied — Hong Kong-listed Hua Hong rose 14% — the market is already pricing this logic. Layer three: China's card is still unplayed. Six rare-earth and related export-control announcements are currently suspended, and the suspension runs to January 10, 2027 — exactly the day the trade truce expires. Layer four: every clock points at the same few weeks — the Miami G20 on December 14–15, the double expiry on January 10, the close of BIS's comment period on January 19, and the license deadline on February 17. The round's biggest hidden loser is not Nvidia; it is the third-country compute middleman — the Megaspeed model turned from a business into a case file overnight.

iHolyshit review

For three days in Shenzhen, the cameras got handshakes and the Federal Register got an effective date. Judge this summit's technology content by two numbers, not by the declaration: how many Section 744.24 licenses have actually been granted by February 17, 2027 — if the answer is zero, "case-by-case review" is a ban under another name — and whether December's action plan in Miami contains a technology annex. If it does not, the tech war keeps running on the BIS track regardless of summitry. My call: licenses will be granted, but only to data centers that have cleanly cut themselves off from Chinese customers — which is precisely the point of the rule: force the world's compute to pick a side first, and do business second.

November 20, 2026

Share this article

What do you think of this story?

Anonymous feedback, up to 5 reactions per article.

Reader comments

Comment without signing in, or sign in to display your member name.

Discuss the technology and the story. Hate, politics and war, threats, abuse, personal accusations, sexual content, ads, links and personal information are not allowed.

0/500

No comments yet. Be the first to share.

Sources