Introduction
The US currently has a lead in AI across chip design, data center build-outs, software tooling, fundamental research, and the production of the best frontier models. However, this lead is fragile.
Chinese firms are rapidly innovating at the technological frontier. At the model layer, DeepSeek R1, Alibaba’s Qwen 2.5 Max, and Tencent’s Hunyuan T1 are now competitive with the best frontier models from OpenAI and Anthropic. At the hardware layer, both Chinese startups such as DeepSeek and tech giants such as SenseTime are buying Huawei’s Ascend 910C chip, which reportedly delivers 60 percent of the performance of an NVIDIA H100 for crucial day-to-day inference tasks. At the application layer, Chinese firms have gone into a frenzy, deploying frontier AI across the value chain, from cars to air conditioners.
To counter this challenge, the United States has relied on export controls to protect its advantage at the frontier. Over the past three years, Chinese access to American AI computing hardware and models themselves has been progressively restricted. These export controls have had the dual effect of constraining China’s ability to produce its own chips and limiting its ability to purchase the most advanced Western AI chips. In turn, these efforts have been repeatedly circumvented by smuggling and front companies, prompting revisions. The culmination of these revisions over the past few years is the Diffusion Rule, which groups the world into three tiers: Tier I for the United States and 18 close allies; Tier III for competitors and adversaries such as China, Russia, and Iran; and Tier II for most other countries in the world. Through its Validated End User Program, the Diffusion Rule ensures American and allied firms remain the partner of choice for building data centers while incentivizing firms in Tier II countries to align with the American AI ecosystem and disentangle from the Chinese ecosystem to receive chips.
However, our advantage will not last forever. In the face of America’s export controls, China has launched a massive bid for self-sufficiency in chip manufacturing, with Beijing taking actions such as launching a $47.5 billion fund for the purpose in May 2024—the third in a series of the semiconductor “Big Fund.” Though self-sufficiency in advanced semiconductors will not happen this year or next, it is certainly possible—and some would say likely—that Chinese firms will eventually build an indigenous manufacturing supply chain producing a competitive volume of AI chips at some point in the future. Once this happens, Chinese firms’ experience reaping efficiency gains at the model layer, as seen in DeepSeek R1, alongside their technical talent, strong hardware manufacturing capacity, and large domestic market, will allow them to rapidly scale compute production, train better models, and build AI data centers across the world, outcompeting America and its allies. By capturing the global market, Chinese firms would reap enormous revenues and data to out-innovate American firms even further, handing China a significant advantage over the United States.
The United States and its allies have a closing window to win on AI. Export controls have bought the United States and its allies time to act to win over the global AI market while we still have advantages in both hardware and software. To complement these export controls, we need to run faster and build our innovation lead.
Therefore, we also need a strategy of “full-stack diffusion” that will durably embed American leadership in AI across the world and widen our gap ahead of China. That means promoting products across our AI stack in a nuanced and strategic manner, using a combination of financing, export controls, standards-setting, and other mechanisms.
