Opinion: Innovation, Not Hardware: What DeepSeek Teaches Us About the AI Race

This week, Chinese startup DeepSeek shocked the world by achieving what many thought impossible: matching U.S. AI capabilities at a mere fraction of the cost. What's more, DeepSeek did this despite U.S. export controls specifically designed to contain China's AI ambitions. While the U.S. has been winning the AI competition with raw computing power, China has found a new way to level the playing field. 

What is DeepSeek, and why is it such a big deal?

DeepSeek has managed to achieve a significant breakthrough in the field of AI. The Chinese startup claims to have trained an AI model that matches or exceeds the capabilities of leading U.S. models—such as OpenAI's GPT-4o—while spending just $5.6 million on computing power. Some experts are skeptical of this number and note that it likely excludes broader research and development costs and potential earlier training attempts. That said, even a significantly higher total cost would still only be a fraction of what U.S. companies spend. For comparison, OpenAI's CEO Sam Altman has suggested that training their latest model cost "well over $100 million."

However, the significance of what DeepSeek has done extends far beyond simple cost savings. The startup has shown that AI development doesn't require the massive computing resources that U.S. tech giants have stockpiled in the last few years. It also shows that China can achieve meaningful AI breakthroughs despite U.S. restrictions on advanced chip exports.

What does this all mean?

On the surface, DeepSeek's success exposes a significant flaw in existing U.S. strategy to slow China's AI development. Washington's export controls on advanced chips were meant to constrain China's AI development, but DeepSeek still achieved its breakthrough. 

DeepSeek's success has shown that U.S. export controls may even strengthen Chinese AI innovation. Chinese companies have been forced to find innovative workarounds to continue AI advancement despite their lack of access to advanced computing chips. While U.S. tech giants have pursued what many call a 'brute force' approach (throwing massive computing resources at AI development), a Chinese company has found a more creative solution. 

China will not roll over and accept defeat simply because of U.S. export controls. They will find a way to pursue AI advancement regardless, and DeepSeek demonstrates that there are paths forward that don't rely on massive computing resources. The question now is whether U.S. companies can adapt their approach as the competition ramps up.

The AI race likely won't be won simply by spending the most money. Instead, it will be determined by who can innovate most efficiently and create advantages that are difficult to replicate.

So what should we do about it?

The U.S. must rethink how it promotes and funds AI research and development. The current approach of backing established tech giants with massive government investments (like the recent $500 billion commitment for AI infrastructure) may be misguided and insufficient to maintain our current lead. DeepSeek's success has shown that finding new and creative approaches to AI development can be far more valuable than simply increasing computing power. 

We must acknowledge that the companies with the most money aren’t always the most innovative. Instead of concentrating resources on a few large tech companies, the U.S. should also seek to promote a greater culture of creativity and innovation through smaller startups. Perhaps then, a U.S. company will produce the next big thing in AI innovation.

Furthermore, we must update our export control strategy. Restricting China's access to advanced chips seemed logical through our previous lens of AI development: "More computing power equals better models. So, let's limit their ability to get more." But DeepSeek proved that hardware limitations alone won't prevent AI advancement. This is not to say we should do away with our current export controls, but rather, we should not view them as the sole solution for limiting China's AI advancement. 

One way we could further slow China's AI advancement is by developing technology that is harder to replicate. DeepSeek and many other Chinese companies use U.S.-made technology and AI models to improve their own. OpenAI has already accused DeepSeek of "inappropriately" taking data from its model. The path forward shouldn't be about winning a computing arms race but fostering innovation that creates durable technological advantages that are difficult for Chinese companies to match. 

We should not view DeepSeek's success as a failure of our own AI development nor as a sign that China has overtaken us. Instead, DeepSeek should serve as both a reminder that China is more than capable of creative innovation regardless of constraints and a wake-up call for U.S. tech companies and policymakers. For us to maintain our current AI leadership, we must realize that our success depends not on outspending the competition but on out-innovating them. Because, as DeepSeek has shown, China can shock the world at any time—and the next breakthrough could put them in the lead…or even end the AI race completely.

The opinions and views expressed in this article are my own and not representative of any company or organization I work for.