The Senate version of the China AI Power Report Act (S.5382) — what it means for the same title to appear in both chambers, and how finely the metrics are specified
A bill requiring a report to Congress on the advanced artificial intelligence capabilities of the People Republic of China, including its efforts on supply chains for advanced AI systems, within 180 days of enactment and annually thereafter for three years. A House bill of the same title (H.R.6275) also exists; twelve such cross-chamber pairs sit among the bills this site holds as of 2026-08-31.
Bill overview (primary data)
- Bill numberS. 5382
- TypeSenate Bill
- Congress119th Congress
- Latest actionRead twice and referred to the Committee on Foreign Relations.（2026-08-07）
Key points
- A bill requiring a report to Congress on the advanced AI capabilities of the PRC and its AI supply chain efforts, within 180 days of enactment and annually for three years.
- Assessment items include total processing power, operations per second, memory capacity and bandwidth, interconnect bandwidth, power efficiency, transistor count and die size, process node, and manufacturing cost and yield assumptions.
- It asks whether a circuit can run models trained on another company hardware, measured in operational terms such as tokens per second and cost per token.
- Specific AI chip design companies are named in the text. At 27,862 characters it is among the longer bill texts this site holds as of 2026-08-31.
- A House bill of the same title (H.R.6275) exists separately; twelve such cross-chamber pairs appear among the 114 bills this site holds as of 2026-08-31.
- At 27,862 characters the text is among the longest this site holds as of 2026-09-01, most of that length from the granularity of reporting items.
1The same title appears in both chambers
This bill carries the same title as H.R.6275, a House bill this site has already covered. In the United States it is unremarkable for identical proposals to be introduced in both the Senate and the House. Whichever chamber advances first, the other falls in behind. Grouping the 114 AI-related bills this site holds as of 2026-08-31 by title yields twelve pairs appearing in both chambers.
Counted one by one they look like separate motions; grouped, they turn out to be the same proposal running twice. Reading bill counts as a count of policies overstates the number.
2What a reporting requirement is trying to measure
Most of the operative text is an enumeration of what to measure and report. Alongside specifications such as total processing power and memory bandwidth, it asks whether a circuit can effectively run models trained on another company hardware — measured in operational terms like tokens per second and cost per token.
Rather than comparing chip performance as a single figure, it reaches for how usable the hardware is when a model is actually running on it. Including manufacturing cost and yield assumptions reads as an attempt to see not only capability but how much can be produced.
3Naming companies in statutory text
The bill names specific companies as objects of assessment. Naming a company in a law that requires a report means the conduct of that company has become a subject of policy judgment. Among the bill texts this site holds, this one is on the longer side at 27,862 characters, and much of that length comes from the granularity of the reporting items.
Writing out what one wants to know at the level of statutory text is the shape legislation takes when the aim is not to regulate but to close an information asymmetry. This bill stands at referral to the Foreign Relations Committee.
4Writing out what you want to know at the level of statutory text
The bill aims not at regulation but at closing an information asymmetry. That character shows in the length of its text and of its title.
Most of the operative text is an enumeration of what to measure and report. Alongside specifications such as total processing power and memory bandwidth, it asks whether a chip can effectively run models trained on another company's hardware — measured in operational terms such as tokens per second and cost per token.
Rather than comparing chip performance as a single figure, it reaches for how usable it proves when a model actually runs, and including assumptions on manufacturing cost and yield reads as looking at how much can be produced.
Why it matters
Counting bills as a count of policies overstates the number by however many run in duplicate across chambers — twelve pairs among the 114 this site holds as of 2026-08-31, so grouping by title is required work. On substance, measuring a chip by whether a model actually runs on it rather than by its specifications is a useful reference for how to compare performance.
FAQ
Why do two bills share a title?
Does this bill regulate anything?
Why measure cost per token?
Sources (primary)
Source: Congress.gov (Library of Congress; U.S. legislative materials, public domain). Links go to the official site.
- Congress.gov (bill page, original)
- S. 5382（119th Congress）