Test report DSG-8527 · Rev F · tested November 18, 2026
Supply Chain & PolicyDevice under test
U.S. Export Controls Now Target Who Uses AI Compute, Not Just Where Chips Go
New U.S. rules require licenses for AI chip exports to five countries and for any data center serving Chinese firms over 12 million TPP, about 760 H100 GPUs.
- Read
- 2 min
- Words
- 497
- Node
- 7nm
- Operator
- Marcus Bennett
Spec summary
- Licenses are now required for advanced AI chip exports to Singapore, Malaysia, Thailand, Indonesia and Brazil.
- A separate control triggers licensing above 12 million TPP, equal to about 760 H100s or 200 B300s.
- U.S. CSPs are exempt from the five-country licenses if exporters report monthly.
- Foundries and OSATs must screen both designers and end users for restricted-country links.
- Chinese CSPs may shift to domestic chips, in-house ASICs or the compliant NVIDIA RTX PRO 5500.

The United States now requires licenses for advanced AI chip exports to Singapore, Malaysia, Thailand, Indonesia and Brazil. Four of the five are in Southeast Asia. A separate control extends further: it requires a license for chips placed in any data center that serves firms headquartered in China or other restricted destinations with more than 12 million TPP of compute. That threshold equals roughly 760 NVIDIA H100 GPUs or 200 B300s.
What does this change for cloud operators?
The rules mark a shift in enforcement logic. Earlier controls tracked where chips physically went. The new framework also tracks who uses the compute, wherever the hardware sits.
Non-U.S. cloud and GPU rental operators serving mainly Chinese customers face the most direct impact. Their business model depends on placing U.S. chips in third-country data centers and renting that capacity across borders. The 12 million TPP threshold now forces a license into that chain.
U.S. cloud service providers sit in a better position. They are exempt from the five-country license requirements, but the exemption carries a condition. Exporters must file monthly reports to keep it.
How might Chinese customers respond?
Chinese cloud service providers have three plausible routes around the new controls, according to TrendForce analysis:
- Domestic Chinese AI chips, produced by local silicon vendors
- In-house ASICs developed by the cloud providers themselves
- Compliant NVIDIA chips such as the RTX PRO 5500, which fall below control thresholds
Each route has a constraint. Supply bottlenecks could limit availability of the alternatives, so substitution will not be instant or complete.
What does the rule mean for foundries and packaging firms?
The controls also reach into chip manufacturing. Foundries and OSATs (outsourced semiconductor assembly and test providers) now need licenses to ship advanced logic chips when the designer is located in, or has an ultimate parent in, the five listed countries or any restricted destination.
This adds a screening burden at two points in the supply chain. Manufacturers must verify both the chip designer and the end user before shipping. A chip that passes the location test can still fail the parent-company test.
Why the five-country list matters
Southeast Asia has become a major transit and hosting region for AI compute. Malaysia, Thailand, Indonesia and Singapore have attracted data center investment as capacity demand grew. Brazil rounds out the list as a non-Asian addition.
The 12 million TPP ceiling gives compliance teams a concrete number to work with. Any data center deployment serving Chinese-headquartered firms above that aggregate threshold triggers the license requirement, and the H100 and B300 equivalents provide a practical conversion for capacity planning.
TrendForce has published a related report, "US Remote-Access Rule Reshapes AI Compute in SEA," covering the market effects in detail.
The combined effect is straightforward. Chip location no longer determines compliance alone. Customer identity and compute volume now sit inside the control perimeter, and monthly reporting keeps exempt U.S. providers tied to the regime.
via insights.trendforce.com (Source)
More from Marcus Bennett
Show full bio
News editor covering regulation, policy and their impact on hardware layer of artificial intelligence at Die Signal.
681 articles
Same lot · LOT-362E
- DSG-99573nmU.S. Closes Overseas Data Center Loophole; Fed Flags $1 Trillion AI Exposure
- DSG-63025nmUS issues interim final rule blocking Chinese remote access to AI servers
- DSG-259945nmNvidia 7%, Apple 15%: Two Paths Into the Same China Trap
- DSG-150745nmUS Tightens AI Chip Exports After Nvidia Smuggling Scandal
- DSG-445745nmU.S. Rule Hits Rented AI Compute: Five Countries, 90 Days