DeepSeek’s Latest Breakthrough Is Redefining AI Race

Photo: Artur Widak/NurPhoto via Getty Images
On January 20, contrary to what export controls promised, Chinese researchers at DeepSeek released a high-performance large language model (LLM)—R1—at a small fraction of OpenAI’s costs, showing how rapidly Beijing can innovate around U.S. hardware restrictions. This launch was not an isolated event. Ahead of the Lunar New Year, three other Chinese labs announced AI models they claimed could match—even surpass—OpenAI’s o1 performance on key benchmarks. These simultaneous releases, likely to be orchestrated by the Chinese government, signaled a potential shift in the global AI landscape, raising questions about the U.S. competitive edge in the AI race. If Washington doesn’t adapt to this new reality, the next Chinese breakthrough could indeed become the Sputnik moment some fear.
News of this breakthrough rattled markets, causing NVIDIA’s stock to dip 17 percent on January 27 amid fears that demand for its high-performance graphics processing units (GPUs)—until now considered essential for training advanced AI—could falter. The performance of these models and coordination of these releases led observers to liken the situation to a “Sputnik moment,” drawing comparisons to the 1957 Soviet satellite launch that shocked the United States due to fears of falling behind.
Until recently, conventional wisdom held that Washington enjoyed a decisive advantage in cutting-edge LLMs in part because U.S. firms could afford massive compute budgets, powered by NVIDIA’s high-performance GPUs. To maintain its edge in the race, the Biden administration implemented export controls to prevent China from acquiring these advanced GPU processors. The release of DeepSeek’s R1, however, calls that assumption into question: Despite limited access to top-tier U.S. chips, Chinese labs appear to be finding new efficiencies that let them produce powerful AI models at lower cost.
Still, upon closer inspection, this falls short of a true Sputnik moment. For one thing, DeepSeek and other Chinese AI models still depend on U.S.-made hardware. Moreover, the AI race is ongoing, and iterative, not a one-shot demonstration of technological supremacy like launching the first satellite. However, the fact that it is not a Sputnik moment should not lull the United States. If the United States does not double down on AI infrastructure, incentivize an open-source environment, and overhaul its export control measures to China, the next Chinese breakthrough may actually become a Sputnik-level event.
The Real Change Is Efficiency
From last month to this month, the real change is the efficiency. DeepSeek researchers found a way to get more computational power from NVIDIA chips, allowing foundational models to be trained with significantly less computational power. Smaller companies and startups will now be able to replicate low-cost algorithms and potentially innovate upon them, enabling the development of more affordable and accessible low-tier and specialized AI applications across various domains.
Of note, China’s sudden leap in AI efficiency highlights the growing impact of open-source collaboration. DeepSeek released R1 under an MIT license, making the model’s “weights” (underlying parameters) publicly available. This move mirrors other open models—Llama, Qwen, Mistral—and contrasts with closed systems like GPT or Claude. In practice, open-source AI frameworks often foster rapid innovation because developers worldwide can inspect, modify, and improve the underlying technology.
From a U.S. perspective, open-source breakthroughs can lower barriers for new entrants, encouraging small startups and research groups that lack massive budgets for proprietary data centers or GPU clusters can build their own models more effectively. Instead of reinventing the wheel from scratch, they can build on proven models at minimal cost, focusing their energy on specialized improvements. That means the next wave of AI applications—particularly smaller, more specialized models—will become more affordable, spurring broader market competition. U.S. companies that embrace these open approaches stand to create robust, adaptable solutions applicable in defense and commercial sectors. Indeed, open-source software—already present in over 96 percent of civil and military codebases—will remain the backbone of next-generation infrastructure for years to come.
If anything, DeepSeek’s accomplishment signals that the demand for powerful GPUs is likely to keep growing in the long term, not shrink. More efficient training techniques could mean more projects entering the market simultaneously, whether from China or the United States. This phenomenon mirrors Jevons’ Paradox: When a resource becomes more efficient, its overall consumption tends to soar. Given the continued importance of U.S.-made hardware within the AI landscape, it’s clear that the demand for powerful GPUs will continue.
Sputnik was a technological feat largely independent of U.S. expertise or resources. During the Cold War, rival powers raced to amass proprietary technologies in near-total secrecy, with victory defined by who could hoard the most advanced hardware and software. The immediate parallel to Sputnik, therefore, overlooks how much of this technology still draws from U.S. research and supply chains. In the AI race, unlike the Cold War, China and the United States draw on each other’s research, open-source tools, and specialized hardware.
China allowing open sourcing of its most advanced model without fear of losing its advantage signals that Beijing understands the logic of AI competition. Each improvement by one player feeds into the next round of global development—even competitors can iterate on publicly shared advances. This leads to faster technology lifecycles and wider adoption, favoring those with vibrant entrepreneurial communities, high-end research labs, and strong venture capital networks. This dynamic, in turn, strengthens the United States’ technology ecosystem by fostering a diverse pipeline of niche AI products, many of which can compete globally.
Long-Term View Instead of a One-Shot Game
First, the Trump administration should adopt a long-term perspective rather than defaulting to retaliatory measures. DeepSeek’s efficiency gains may have startled markets, but if Washington doubles down on AI incentives, it can solidify the United States’ advantage. This means investing not only in ambitious programs targeting advanced AI (such as AGI) but also in “low-tier” applications—where high-volume, user-focused tools stand to make an immediate impact on both consumers and businesses.
Open-source AI development is key to this strategy. Open-source projects allow smaller startups and research teams to participate in cutting-edge work without massive budgets. To bolster this trend, the White House could offer tax credits or accelerated depreciation for private-sector investments in open-source AI. Such policies would also encourage deeper collaboration with allies and partners, harnessing the United States’ vibrant entrepreneurial culture and extensive research network.
Second, the export-control measures must be rethought in light of this new competitive landscape. DeepSeek’s breakthrough underscores that the AI race is continuous, the gap between the United States and China is narrower than previously assumed, and that innovation by industry startups is the backbone of this race. U.S. strategy of containment with export controls will surely limit the scalability of the AI industry within China. Even though DeepSeek’s R1 reduces training costs, text and image generation (inference) still use significant computational power. This will likely be a bottleneck, preventing China from scaling its AI service offerings to the globe, under tightening chip sanctions. However, the DeepSeek example showed that export controls cannot kill innovation. Those who are not able to access these chips will innovate their own ways. These blanket restrictions should give way to more detailed and targeted export-control systems. Recent AI diffusion rule puts 150 countries in the middle tier category in which exports of advanced chips to these countries will face difficulties. Those countries will either innovate their own industries or will develop ties with China. This is why such a blanket approach will need to be reconsidered. Moreover, U.S. export control policies must be paired with better enforcement to curb the black market for banned AI chips.
Finally, the Trump administration should invest in robust evaluation programs to identify and mitigate bias in emerging AI models. As smaller, specialized applications gain traction, transparent testing frameworks become vital for building public trust and ensuring market scalability. A system that flags and corrects issues—like DeepSeek’s purported bias on China-related topics—can ensure these models remain globally relevant, fueling further innovation and investment in U.S.-led AI research.
Ultimately, DeepSeek is not a Sputnik moment, yet. If the United States adopts a long-term view and strengthens its own AI eco-system encouraging open collaboration, investing in critical infrastructure, it can prevent a Sputnik moment in this competition. DeepSeek is indeed a boon for the AI industry. By adopting these measures, the United States can increase its share significantly in this growing industry.
Yasir Atalan is a data fellow for Futures Lab in the International Security Program at the Center for Strategic and International Studies in Washington, D.C.