A Chinese company’s latest artificial intelligence (AI) technology has sent shock waves through the U.S. AI sector. In the past few years, U.S. companies OpenAI, Google, Meta and Anthropic have dominated leaderboards for cutting-edge AI with Microsoft, Amazon and Nvidia leading in cloud and specialized AI hardware. To play in this league requires a huge war chest of billions of dollars.
Now, DeepSeek, the AI division of a Chinese hedge fund and a (relatively) much smaller company, has released large-language and reasoning AI models whose performance is comparable to the best-in-class U.S. AI models. According to the company, it did not use high-end GPUs (i.e. AI chips), and the computing bill was estimated to be only $5.6 million, which is a fraction of what it costs rivals to build their current models. Expert and industry reactions to these claims ranged from skepticism to shock and awe. Based on the details in the technical reports from DeepSeek, this breakthrough, prima facie, seems plausible.
Credit: Handout
Credit: Handout
Credit: Handout
Credit: Handout
Making available such an advanced AI model with a permissive open source license (i.e. almost giving it away for free) throws a wrench into the pricing strategies of commercial AI vendors who have to recoup their massive up-front investments. It might also mess up many 401(k) plans stateside.
For the Chinese government, DeepSeek is an inspiring David to U.S. Big Tech’s Goliath: local teams educated at top Chinese universities innovating against odds, delivering step-change cost reduction, open sourcing the latest AI models, and publishing their methods for others to leverage. Don’t be surprised if such open source AI is branded as a digital Belt and Road Initiative and a tool of Chinese soft power: AI solutions for the world, including the Global South, at a fraction of the cost.
To be clear, this development does not dethrone the United States as the global leader in AI but narrows the gap between the United States and China; it requires the U.S. government to urgently reassess AI policies. It poses serious questions to U.S. AI companies about their competitive moats and the economics of commercial AI and requires them to refocus strategies and investments. These questions are important not only for policy wonks, tech moguls and venture capitalists; they are also relevant for everyday Americans whose 401(k) accounts could take a big hit if American AI leadership is not maintained. Consider that 62% of adult Americans (including 80% of older Americans) invest in stocks. The S&P 500 returned 23.3% in 2024. Without the “Magnificent Seven” technology companies (Alphabet, Amazon, Apple, Meta, Tesla, Microsoft and Nvidia, all significant AI players) — it would have been only 6.3%. It is not surprising that around the world the initial stock market reaction has been brutal.
In response to this disruption, there are a few initiatives on which the U.S. government and industry leaders can focus. The U.S. government, private equity, venture capital and other investors are pouring billions of dollars into AI. The Trump administration brought SoftBank, Oracle and OpenAI together for an event to announce a $100 billion investment in AI computing centers (code-named “Stargate”). The scale of these investments might need recalibration. There are many ways to make industry-friendly regulation here (construction, real estate, energy policies).
Notwithstanding these big initiatives, much policymaking has been left to the courts. High-stakes lawsuits are snaking their way around the judiciary. Before these cases are decided, the U.S. government should provide clarity on regulation at federal level. The United States also needs to resist stamping out open source AI. If the government does that, it risks ceding space to China and others. Let us not forget that open source lets U.S. AI companies take advantage of DeepSeek’s new advances in this round.
Next, there is the issue whether U.S. export controls are serving their purpose. Indeed, Nvidia sells the H800 chip, which are the modified (slowed down) versions of the H100 chips that Washington has restricted Nvidia from selling to China. There are news reports of reexports and smuggling of H100s to China, something that is sure to concern the U.S. national security complex. The U.S. government must assess whether another round of export controls on the best semiconductor chips will be efficacious. Better enforcement against smuggling of GPUs might be in order. The Trump administration needs to rethink export controls. At the very least, it needs to monitor and enforce better. Ironically, AI will be helpful here. Also on the to do list is a rethink of regulation based on amount of compute used to train a model.
Generative AI, while important, is only one tool in the AI solutions portfolio. Generative AI is crowding out other AI investments, such as biomechanics and robotics. By leveraging U.S. research institutions and academic centers, these areas can be prioritized and grown. A talent strategy, even one encompassing immigration issues, needs to be seriously considered as well. David Sacks, the new AI czar, and the Trump administration have their work cut out for them. They should also consider devising an interim plan while fleshing out a longer-term vision that was mandated last week in an executive order.
The U.S.’ AI leadership position needs to be protected, not just because AI is transformative for the future or geopolitically critical. The financial health of the American AI industry and Americans is at stake.
Kashyap Kompella is founder of RPA2AI Research, an AI analyst firm. James Cooper is a law professor at California Western School of Law in San Diego. They are coauthors of “A Short and Happy Guide to Artificial Intelligence for Lawyers.”
