January 12, 2026
America will only win the AI race by ensuring our AI stack becomes the global standard. By contrast allowing a parallel AI ecosystem to proliferate will only hurt our ability to dictate standards and protocols. The administration’s decision to allow licensed sales of NVIDIA’s H200 chips to vetted commercial customers in China is not a retreat from U.S. leadership but is an America-first strategy to ensure that our system is the one that dominates the market. It is also a return to a sliding-scale for export controls that ensured American technology dominance since the birth of the modern computer industry.
The primary objective in the AI race is not to block every foreign application, but is to ensure the American AI stack becomes the global default. This can be accomplished while preserving a decisive technological lead for U.S. firms and allies. We have real proof of this with 5G where export controls enabled Huawei to become the primary partner for much of the world.
Blanket export bans have repeatedly backfired. When U.S. suppliers are forced out of large commercial markets, Chinese firms do not pause but find substitutes. Too often overly broad controls have unfortunately only accelerated the rise of domestic Chinese alternatives. Time and time again we have seen this as Chinese developers build onto non-U.S. stacks they otherwise would not have chosen. This dynamic ultimately undermines U.S. influence, fragments global standards, and hands long-term ecosystem control to our competitors.
Allowing controlled exports of H200 that are one—and soon two—generations behind the latest versions avoids that outcome. This approach allows the US to retain our edge. American hyperscalers are already deploying Blackwell, all while NVIDIA has unveiled its next-generation Rubin architecture. AI systems improve at exponential rates. The newest generation launching this year is more than 20 times more efficient than chips from 2023 or 2024. Selling prior-generation accelerators does not erode America’s lead but preserves it by keeping global developers anchored to U.S. hardware and software frameworks.
Critically, the administration approved sales are limited to purely commercial end users that are not on entity lists. To be clear if these companies posed national-security risks, they would already be prohibited from buying U.S. CPUs, software, and cloud services. Additionally, Chinese military and sensitive government uses are already required to rely on Chinese domestic chips.
Instead of blanket bans, national security is best protected by targeted controls specifically end-use restrictions, licensing, monitoring, and enforcement. We are not protected by forfeiting entire markets. A calibrated “sliding scale” approach has long been standard U.S. practice: protect the frontier, permit legitimate commercial exports, and prevent diversion.
The Administration is leading, and Congress must meet the moment. Lawmakers should put forward legislation that sets the global standard for American innovation, not policies that slow it down. Measures like AI Overwatch wrongly frame American technology as a weapon rather than recognizing it as the backbone of our future economic security.
The AI race will be decided by which country’s technology stack becomes the global standard. Instituting policies that enable U.S. platforms to be embedded worldwide will only strengthen American leadership and reduce reliance on our competitors.
.png)
