Selling semiconductor equipment allows China to create hundreds of billions in downstream value. In contrast, selling API access to US models is a higher-margin strategy that keeps core value creation within the American ecosystem, extracting more revenue per unit of capability provided.
Related Insights
A proposed policy for China involves renting access to US-controlled chips (e.g., in Malaysian data centers) instead of selling them outright. This allows Chinese companies to benefit commercially while giving the US the ability to "turn off" the chips if they are misused for military purposes.
The pro-export argument for selling NVIDIA chips to China is strategic: flooding their market with good-enough, affordable chips makes it uneconomical for their domestic industry to compete. This fosters dependency on the U.S. ecosystem and can slow their independent technological progress.
By releasing powerful, open-source AI models, China may be strategically commoditizing software. This undermines the primary advantage of US tech giants like Microsoft and Google, while bolstering China's own dominance in hardware manufacturing and robotics.
The decision to allow NVIDIA to sell powerful AI chips to China has a counterintuitive goal. The administration believes that by supplying China, it can "take the air out" of the country's own efforts to build a self-sufficient AI chip ecosystem, thereby hindering domestic firms like Huawei.
Allowing H200 chip sales gives China significant AI compute capability. This short-term revenue boost for NVIDIA won't alter China's long-term policy of reducing reliance on foreign tech, effectively helping a competitor in a strategic race.
Unable to compete globally on inference-as-a-service due to US chip sanctions, China has pivoted to releasing top-tier open-source models. This serves as a powerful soft power play, appealing to other nations and building a technological sphere of influence independent of the US.
A nation's advantage is its "intelligent capital stock": its total GPU compute power multiplied by the quality of its AI models. This explains the US restricting GPU sales to China, which counters by excelling in open-source models to close the gap.
While the U.S. AI strategy pursues a 'winner-take-all' model leading to high profits, China's state-backed approach aims to commoditize AI. By spreading resources across many players to create a low-cost, replicable model for export, it structurally limits the potential for monopoly profits to accrue to shareholders.
U.S. chip companies that sell to Chinese tech giants are making a strategic error. They are building a temporary bridge for future competitors who are mandated to switch to domestic suppliers like Huawei once viable. This short-term revenue comes at the cost of shrinking their own long-term global market share.
A complete ban on selling chips to China is counterproductive. The ideal policy allows NVIDIA to sell chips that are one or two generations behind state-of-the-art. This strategy keeps Chinese firms dependent on the NVIDIA ecosystem, funds U.S. R&D with sales revenue, and hinders domestic competitors like Huawei from flourishing.
