Plan A would require AI chip factories, buyers and data centers to register with governments and undergo inspections so the US and China can track advanced chips and jointly enforce limits.
The proposal says those controls would raise chip costs only by a few percent, while targeting high-end hardware such as Nvidia’s roughly $40,000 H100 rather than ordinary laptops or phones.
Consumer devices would face limits only if hardware output surged enough to make distributed training viable; Plan A estimates training one frontier model would need 5%-10% of all current phones and computers.
After 2030, the framework would ban training new open-weight models while requiring open algorithms instead, aiming to spread know-how across about 10 firms in 3-6 countries by 2035.
The plan argues these measures are standard industrial regulation, not mass surveillance, and says similar chip-customer KYC rules were already enacted by the Trump administration in January.