Winning the AI race means beating China on innovation, cost and trust

Because we operate at the intersection of AI, national security and cyber, I often get asked the question, "Are we winning the AI race with China?"

My answer is still yes. At the highest level, we have the best chip technology, the best models and, more importantly, we have the best economic system and the talent to continue to win.

But at a deeper level, we still have to agree on what winning looks like. Fundamentally, is it building the best technology, or is it building the technology that becomes ubiquitous?

CRUZ SAYS WINNING AI RACE AGAINST CHINA IS 'SINGLE MOST IMPORTANT ECONOMIC QUESTION' FOR AMERICA

The tech world is littered with examples of superior technologies that lost to inferior ones that took over the market and became the standard. Decades ago, when the stakes were much lower, VHS beat Betamax. More recently, and much more relevant to our national security, Huawei beat its Western Hemisphere competitors to become the global leader in telecommunications gear. This handed China a major opportunity to gather information, intelligence, and leverage around the world.

So, the more complete answer on the race with China is that it is a race for global adoption, not just for technology superiority. When the dust settles and the picture becomes clearer, what will matter most is whose technology has become the global standard. Which AI stacks are people around the world using to become informed, to automate, to increase productivity, to analyze and decide?

BEYOND KILLER ROBOTS: CHINA'S REAL AI THREAT TO AMERICA IS ALREADY HERE, EXPERTS WARN

Within the global AI adoption race, the contest is complex and better described as a triathlon, with all three legs being run simultaneously.

The first leg is the innovation race, where America is well ahead. Experts estimate we have at least a two-year lead on chips thanks to our fundamental advantage in lithography and incredible advances by Nvidia, its Taiwanese partner TSMC and many others. And on the models themselves, while the gap is narrowing, our frontier labs are leading by perhaps two to three generations — so eight to 12 months. That may feel short, but it is a lifetime in frontier AI.

US ADVERSARY TURNS UP THE HEAT ON AMERICA’S AI LEAD WITH AN UNDERESTIMATED EDGE

Take the example of AI in cybersecurity, which has rightly been much in the news. Booz Allen’s Cyber Weapon Index codifies how good these models are at conducting a cyberattack. Two models, Mythos from Anthropic and Astra from OpenAI, outscored all others by a wide margin. This is good news because both companies are speaking publicly about the need to behave responsibly and in partnership with the U.S. government in releasing these technologies to the world. But several Chinese models have shown initial capability and are growing in expertise — getting better in every generation — likely with fewer guardrails than their American counterparts.

The second leg of the AI adoption triathlon is based on cost. Frontier AI models are powerful but not cheap. Roughly speaking, the difference between the best American models and the best Chinese models is 5-10x the cost per token. And while the Chinese models cannot fully replicate the capabilities of our frontier labs, they are often good enough for many tasks. As a result, they are being used widely, especially by cost-sensitive customers. Think large global companies trying to manage IT budgets, cash-strapped startups in Silicon Valley, and developing-country governments with limited resources.

According to OpenRouter, a model marketplace where users can access a range of models, roughly 50% of tokens used in the last year were consumed on Chinese models. And we know of many U.S. startups that are using Chinese models — sometimes not realizing or disclosing their actual provenance — as code assistants or as the substrate for their applications. This matters because Booz Allen’s research demonstrates that Chinese models introduce more vulnerabilities when they code for an American application than when they do for a Chinese one. Over time, the accumulation of these tiny vulnerabilities could undermine the entire software supply chain on which the U.S. economic future rests.

And the third leg of the AI adoption triathlon is trust. Companies, governments, and individuals are less likely to use a technology they can’t control or believe is operating against their interests. Here America — based on our values, way of life, free-market system, and history — has a right to win. When American ingenuity created the internet, for example, the world eagerly adopted it, in no small part because the decentralized, transparent system of rules made it easy to use, to understand and ultimately to trust. By contrast, the internet as it exists behind China’s Great Firewall, with its more rigid state controls and pervasive surveillance, would not be a good fit for most democracies around the world.

CHINA ‘THROWING GASOLINE ON THE FIRE’ IN AI DATA CENTERS RACE, SAY LAWMAKERS

Despite the underlying American advantage, the trust race is much closer than it should be. Both countries and their respective AI ecosystems are eroding necessary trust unnecessarily. China, for example, trains its models to refuse to answer questions that go against Communist Party dogma, and to even refuse to perform certain tasks that the model perceives as being contrary to CCP interests. In America, for many reasons, including disinformation and the lack of a clear framework, polls suggest our fellow citizens are turning negative on broad issues from the construction of data centers to the speed of AI advances.

America must win this adoption triathlon by simultaneously pushing on all three fronts. Winning is key to our national security, our economic security and our global standing. We must continue our leadership on the actual technology stack, invest in lower-cost alternatives to the frontier models and rebuild trust. The president’s America’s AI Action Plan provides a roadmap. A few ideas to strengthen it would include:

DATA CENTERS COULD BE THE SLEEPER ISSUE OF THE 2026 MIDTERM ELECTIONS

And move fast. By one measure, AI models now double their capability every four months. A good goal would be to have a critical infrastructure designation, framework and communication mechanism by the end of 2026 — before models are so advanced that they are designing themselves (aka recursive self-improvement, or RSI), or we have slowed ourselves down so much that the Chinese ecosystem is well on its way to global adoption.

The future has arrived. Let’s widen our lead.