The rule licenses shipments, not logins
Under current law, cloud computing as a service is not an export, so BIS cannot license a login directly; the rule works around that gap. It attaches the license requirement to the chips and servers themselves: if an exporter knows the hardware is headed for a data center that will give covered companies access above the threshold, the shipment needs a license. A later switch to restricted use is treated as an in-country transfer that also needs authorization.
That design matters for who carries the burden. Exporters have to be able to show what they knew about a data center's tenants. According to TechTimes' account of the rule's Note 3, a foreign data center that cannot supply enough ownership information is presumed to give the exporter that knowledge unless the red flag is resolved first — which in practice pushes ownership and use attestations down onto tenants, as BERI's analysis of the rule describes. BIS's reported paperwork estimate is roughly 100 additional license applications a year; that figure does not itself quantify all the compliance work underneath it.
Read the 750-chip figure as an estimate
The threshold is 12 million aggregate TPP, where TPP is BIS's total processing performance measure: twice a chip's theoretical peak multiply-accumulate rate, times bit length, aggregated across the facility or linked facilities. TechCrunch estimated that the line works out to roughly 750 Nvidia H100 or H200 chips. That equivalence is TechCrunch's approximation, not a number in the rule, and it moves with the chip: newer, faster parts reach 12 million TPP in fewer units, and the total counts across linked sites rather than per building.
Two scope details decide more cases than the headline number. First, the count aggregates compute provided to covered companies, so several smaller tenants can push a site over the line together. Second, the five-country license requirement is separate from the tenant test: a shipment to Singapore or Brazil can need a license on destination alone, unless the recipient is headquartered in, or ultimately parented in, the United States and the reporting condition is met.
Announced is not in force
BIS posted the text on November 16 and it went on Federal Register public inspection the same day. By its own terms it takes effect on publication; a compliance guide gives November 19, 2026 as the scheduled publication date, which ALFW had not independently confirmed when checked. These new provisions are not effective before publication; existing controls remain applicable. Some reports described the rule as taking effect in 90 days; the 90 days in the text is the window existing operators get, after the effective date, to obtain licenses for customers they already serve. The 60-day comment period also runs from publication, not from the announcement.
The rule also formally rescinds the Biden administration's January 2025 AI Diffusion Rule. That framework set country tiers for chip exports and, by BIS's own account, was never enforced. Replacing a never-enforced tier system with a shipment-and-knowledge test is a narrower instrument, aimed at a specific route: renting compute in a third country instead of importing the chips.
What changes for operators and renters
For data center operators in the five named countries, and anywhere else serving D:5- or Macau-headquartered tenants, the practical change is documentation before it is prohibition. Tenants should expect to be asked who owns them, who their ultimate parent is, and what the capacity will be used for, with a duty to flag changes. Reuters reported Chinese AI companies rushing to secure overseas computing capacity in anticipation of tighter restrictions; for firms in that position, the question is no longer only where the servers sit. It is whether the shipment that stocked them was licensed for that use.
None of the five named governments had issued a detailed response in the coverage reviewed, beyond Singapore's joint statement that it is studying the measure. The first real test dates are publication (scheduled for November 19) and Nvidia's earnings on November 17, where licensing effects in the five countries are a plausible question, though the rule will not yet be in force when Nvidia reports.
Sources
- [1] MDDI–METI Joint Media Statement on US Interim Final Rule on Addressing Remote Access to Advanced Semiconductors, Singapore, November 16, 2026 (primary confirmation of the five-country scope and the US-headquartered exception; read first-hand by ALFW)Read source
- [2] TechTimes, "AI Chip Export Rule Sets Cloud License Line at Roughly 750 Nvidia H100s for China-Linked Tenants", November 16, 2026 (account quoting and linking the BIS rule text throughout; read first-hand by ALFW)Read source
- [3] BIS press release, "BIS Issues Rule Addressing Remote Access to American Super Intelligence Technology", November 16, 2026 — via bis.gov (label only; URL not confirmed by ALFW)
- [4] Interim final rule text, "Addressing Remote Access to Advanced Semiconductors", RIN 0694-AK50, BIS, November 16, 2026 — via bis.gov (label only; RIN, section number and rescission details rest on this text)
- [5] Superintelligence Compliance, GAIN AI Act and AI Chip Export Controls guide (November 19 scheduled publication date; public-inspection status)Read source