President Trump announced Monday that Nvidia will pay the U.S. government 15 percent of the revenue it derives from sales of its H20 chip to China—an unusual arrangement that gives up leverage. The United States could do better: Rather than selling AI chips, the United States should instead rent them to China to permanently maintain its AI chip edge (PDF).
Controversy surrounding NVIDIA's China-specific H20 AI chip has reignited debate over chip exports to China. Proponents of H20 exports argue that sales keep China dependent on U.S. technology while generating revenue. They claim denying access would only accelerate China's drive for homegrown alternatives. These arguments miss a superior alternative to selling chips that would yield similar benefits to the United States: providing China with the computing power it seeks for a fee via remote cloud access
When the United States sells chips, they're gone forever—disappearing into data centers with unknown users pursuing unknown goals. There's another option: give users the computing power of chips remotely.
When the United States sells chips, they're gone forever—disappearing into data centers with unknown users pursuing unknown goals. There's another option: give users the computing power of chips remotely. Cloud services offer almost everything chip sales promise, plus actual control: access can be shut off at any stage (PDF) as geopolitical conditions change or threats arise. The economics are better too. AI chips are sold to U.S. cloud providers, who generate additional revenue by renting AI chips to China—creating two revenue streams instead of one. Companies like Amazon and Microsoft already serve global customers this way, maintaining physical possession of the chips while providing computational power on demand.
Unlike physical chip exports, cloud access makes it significantly harder for customers to redirect access to unauthorized parties. Know-your-customer requirements enable monitoring for connections to Chinese military entities. Cloud access also creates continuous leverage—every renewal becomes a negotiation opportunity, compared to chips that get replaced only every three to four years.
Realistically, neither chip sales nor cloud access will stop China's semiconductor sovereignty efforts, regardless of the fact that they lag far behind the United States and partners' capabilities. China long ago committed to pursuing chip independence as a strategic imperative, not an economic calculation, as it never perceived the United States as a reliable supplier. Whether or not Chinese firms buy or rent U.S. chips, Chinese chip fabs will continue advancing, and Huawei will keep developing its Ascend AI chips.
Realistically, neither chip sales nor cloud access will stop China's semiconductor sovereignty efforts, regardless of the fact that they lag far behind the United States and partners' capabilities.
Moreover, Beijing appears to be using regulatory uncertainty to create a captive market that would match Huawei's limited production. It has raised concerns that Nvidia's chips may contain “backdoor” functions that could allow remote access to chips after acquisition. Beijing has also introduced energy rules that, if implemented strictly, could prevent Chinese companies from using H20 chips. These efforts aim to preserve sufficient demand for China's limited domestic chip production, even when more advanced U.S. chips are available.
If the United States wants to make a deal on AI chips, it could provide them through the cloud to maintain control, generate recurring income, and preserve strategic flexibility. Smart landlords don't sell their buildings for quick cash—they rent them. When tenants misbehave, you can kick them out and still own the asset. That's real leverage. Selling AI chips to China outright reduces America's AI lead for little benefit. So rather than exporting world-class AI chips, the United States should rent them out.
