America Is Winning the AI Race—But We Can’t Afford to Act Like It
Here’s the thing you need to understand about the artificial intelligence race: it’s exponential, not linear. In a traditional race, the track underneath you doesn’t change in real time. But when it comes to AI, a single innovation can radically transform the field. Think of a poker table, where the deck reshuffles mid-hand.
In January, a little-known Chinese startup called DeepSeek jolted the stock market with a cheaper reasoning model than anyone thought possible. Immediately, AI chip and Big Tech stocks plummeted, the Nasdaq slid about 3%, and Nvidia shed roughly $590 billion in market value—the largest one-day wipeout on record.
Last month, Nvidia published a Nature paper, claiming it trained its flagship “R1” for about $294,000 on export-restricted H800 chips. That’s a suspiciously low number, to say the least. For context, the H800 is a purposely less capable chip in order to get around export controls. Most in the field are quick to dismiss this problem, claiming the U.S. is still well ahead of China. But it’s still prudent to take into account the existence of a competitive, cheap, Chinese model.
With that said, are we winning the AI race? Absolutely, and it’s not particularly close. U.S. institutions produced 40 notable AI models in 2024, compared to China’s 15. Private U.S. AI investment hit roughly $109 billion last year, nearly 12 times China’s $9.3 billion. The U.S. currently holds 75% of global compute capacity, compared to China’s 15%. Capital and compute power have converged in one place: the United States. We ought to keep it that way.
Even as the U.S. leads on most headline metrics, there’s an important trend we can’t ignore. Since DeepSeek’s release, Chinese teams have been producing AI models that are competitive with some of America’s—even when relying on lower-end chips. When the U.S. first rolled out its early-generation models under similar hardware constraints, they were arguably less capable than what China is producing today. Put differently, China’s ability to wring high performance out of limited compute is striking. We may still dominate in aggregate compute, capital, and scale—but that dominance should not lull us into complacency.
Some of China’s models are highly competitive in domains like coding and image generation. U.S. firms may still hold the edge in quality, but China’s ability to deliver comparable performance at far lower cost is an advantage in its own right—one that accelerates adoption and subtly erodes the certainty of U.S. leadership.
Cut off from top-shelf chips by U.S. export controls, Chinese teams have learned to squeeze more out of what they have. In that way, constraints can be catalysts for innovation. America’s most concrete advantage is still compute, but if abundance dulls our drive to innovate on efficiency, or if overregulation slows deployment as it has in Europe, that edge could narrow.
If the goal is not just to win but to dominate, we must beat China at its own game, not only the one they are barred from playing.
What makes the AI race with China even more unpredictable is the Chinese Communist Party’s stringent approach to information control. Over the last five years, Beijing has expanded its reach into nearly every sector, including technology and finance. This typifies the environment in which Chinese capabilities develop: opaque, highly centralized, and meticulously shielded from outside scrutiny. The revised State Secrets Law, along with the Cybersecurity Law, broadens the definition of sensitive information to include any data, documents, or activities Xi Jinping considers “national security,” effectively criminalizing much of what foreign analysts might try to measure.
As a result, intelligence asymmetry is built in: China knows more about us than we know about them. That means they have the capacity to surprise us. The CCP’s warning shots against firms like Mintz, whose Beijing office was raided, and Bain & Company, whose staff were questioned by authorities, illustrate just how determined Beijing is to ensure that the U.S. maintains only a limited grasp of Chinese developments. It is only reasonable to conclude that China wants us overconfident. Indeed, their self-perception as a rising yet quiet giant makes this expectation central to their strategy.
In light of China’s proven ability to innovate and have tight controls on information, the only safe margin is a widening one. DeepSeek reminded markets how quickly the deck can reshuffle. Our job is to ensure we have enough of a lead to offset any surprise cards Xi may be hiding up his sleeve. We will not achieve that by resting on our laurels, but by continuing to push the boundaries. In this race, anything short of dominance is defeat—for it is more than an arms race. It is a civilizational contest.
The post America Is Winning the AI Race—But We Can’t Afford to Act Like It appeared first on The American Mind.