The Chinese start-up has blown up our assumptions about American tech supremacy—and made the AI race very real.

By Geoffrey Cain
Published January 28, 2025

On Monday, Nvidia’s stock price fell by 17 percent. The firm, whose chips have powered America’s AI revolution, lost nearly $600 billion in market capitalization—the largest single-day sell-off in history. Other tech stocks plummeted too, with more than $1 trillion wiped off U.S. stock markets.

Why? Because DeepSeek, a small Chinese start-up almost no one had heard of until last week, announced that it had built high-performing AI models at a fraction of what America’s tech giants spend on theirs. The firm claims that it took just $6 million and two months to build its reasoning model, DeepSeek-R1. It works at least as well as its American rivals’ models, which required billions of dollars’ worth of state-of-the-art chips, and DeepSeek’s software has jumped to the top of the app store sales charts, leapfrogging OpenAI’s ChatGPT.

By delivering high-performance capabilities without demanding the staggering computational resources of Western AI models, DeepSeek brings cutting-edge AI within reach of anyone with modest resources. In doing so, it has blown up our assumptions about American supremacy in the AI race—and could herald a period of dangerous instability.

While DeepSeek’s low-cost breakthrough has upended the AI world, the disruption is more profound than that.

A cheap-to-make, Chinese-owned AI model could be the harbinger of a frightening new reality. Think of the way in which cheap drones have transformed the modern battlefield, as demonstrated by the war in Ukraine. Now imagine swarms of low-cost drones deployed in war zones, powered by low-priced DeepSeek-style AI, able to coordinate attacks, evade defenses, and overwhelm traditional tanks and infantry platoons, all with unprecedented precision and scale. Military strategies that once relied on large arsenals and expensive hardware are vulnerable to disruption by nimble, cost-effective alternatives.

The unsettling truth exposed by the DeepSeek breakthrough is this: The traditional protections of industrial and military supremacy are far weaker than we thought.

For decades, the United States has counted on the fact that it held an unshakable technological edge. More recently, Silicon Valley’s giants, with their proprietary AI models and sprawling research labs, thought themselves invincible. Huge capital costs—along with a seemingly ironclad U.S. sanctions regime against China—would serve as a deep moat around America’s AI advantage. DeepSeek called all this into question and exposed the complacency of America’s tech industry and policymakers.

How did this happen? Part of the answer is that America’s leaders gave short shrift to the Chinese approach to innovation, a dog-eat-dog cage fight that prizes speedy imitation over risky, colossal investments at the bleeding edge. Chinese firms are more interested in being fast followers than first-movers.

“It doesn’t matter where an idea comes from or who came up with it,” argues Taiwanese AI expert Kai-Fu Lee, whose firm Sinovation Ventures backs AI start-ups in China, in his book AI Superpowers. “All that matters is whether you can execute it to make a financial profit.”

Since the Second World War, the U.S. has been the world’s technology first-mover, pioneering the breakthroughs in semiconductors, space exploration, and computing that have driven social change and economic growth. China, playing catch-up, has mastered the art of scaling and refining these innovations. China’s companies often take a systematic approach to transferring, even copying, foreign intellectual property, a strategy of government-supported, industrial-scale espionage.

This was exactly the dynamic of Apple’s failed ambitions to build a long-range electric vehicle (EV) battery in partnership with Chinese automaker BYD. The project had the potential to expand the life of EV batteries everywhere. But last October, after seven years of research and development, Apple abandoned the project, while its partner BYD surged ahead with similar technology. BYD has now overtaken Tesla as the world’s largest maker of electric vehicles.

The sad stretch of complacency that has brought us to this Sputnik moment is reflected in how we train the next generation of technologists. China produces nearly five times more STEM graduates each year than the U.S., according to the National Science Foundation. In 2022, China accounted for 31 percent of global manufacturing output—the largest share in the world—and more than double the United States.

And our efforts to reshore much-needed chip manufacturing industries have been underwhelming. Under the Biden administration, a massive industrial project called the CHIPS and Science Act was supposed to boost our chip-making and secure the future. But the numbers involved weren’t as impressive as they may have seemed: While the entire federal government allocated $52.7 billion to these chip projects, a single company in Taiwan, TSMC, expects to spend up to $42 billion this year alone.

Meanwhile, American companies’ presence in Chinese markets means we have given away technological secrets that have major national security implications. Microsoft, to name just one firm of many, has inadvertently accelerated China’s AI growth by providing the country with lucrative research partnerships, cloud computing services, and infrastructure.

DeepSeek’s announcement should be a wake-up call. But what can we do to reclaim our edge?

The lesson from DeepSeek’s low-cost breakthrough isn’t that we should avoid investing in expensive chips. Quite the opposite. We’ll need to invest more if we hope to win the AI race. We shouldn’t move forward on the assumption that the DeepSeek model of fewer chips will work for every other innovation. And if we miss one generation of chip-making, it may be impossible to catch up.

Future AI applications—in our self-driving cars, medical equipment, and military weaponry—will become ever more advanced and will inevitably need more computing power, which could arrive in the form of the innovation called the “chip stacking method” or quantum computing.

And so we must reinvigorate the art of thinking big that drove our ancestors who engineered the Apollo moon landing and built the transcontinental railroads. Large-scale investments will have to be critical. Project Stargate—the Trump administration’s $500 billion venture with SoftBank, OpenAI, Oracle, and others, designed to keep the U.S. at the forefront of AI—is a good start.

Second, chip manufacturing and AI development must be treated as a national security project, not only a profit-making one. Without a reliable domestic chip supply, America’s technological dominance will continue to be eroded.

Finally, the U.S. has to stop the damaging technology transfers by Big Tech companies too eager to do business with China. American innovations must remain American; American companies must show their allegiance to one flag rather than being doublehanded with enemies.

China’s unexpected AI leap forward may have been unwelcome news—but the really dangerous thing would be to ignore this wake-up call. Recent AI advances on American soil have lulled us into a false sense of security. DeepSeek must do the opposite.

See Also:

• DeepSeek hired talent from Microsoft’s controversial AI research lab in China
• WATCH: Can America Win the AI War with China?