US Export Controls May Have Unintentionally Boosted DeepSeek's AI Growth, Raising Concerns in Washington

Tight US export restrictions on advanced chips appear to have inadvertently fueled the rise of Chinese start-up DeepSeek, intensifying fears in Washington about America’s ability to maintain its lead in the global AI race.

Based in Hangzhou, DeepSeek has surprised investors and industry experts with its R1 chatbot, which rivals American counterparts at a significantly lower cost. This achievement comes despite stringent US rules limiting China’s access to high-end chips vital for training large AI models.

DeepSeek’s founder, Liang Wenfeng, acknowledged that the “embargo on advanced chips” posed major challenges. However, while these restrictions were designed to safeguard US technological dominance, experts believe they may have encouraged DeepSeek to find creative workarounds.

The company utilized H800 chips—less powerful but legal for export to China until late 2023—to develop its large-scale AI model.

“Restrictions on China’s chip access pushed DeepSeek’s team to create more efficient models that perform well without requiring massive computing resources,” said Jeffrey Ding from George Washington University.

Ding added that DeepSeek’s progress demonstrates that “US export controls are ineffective at halting other nations from advancing cutting-edge AI models.” He noted that history shows it’s nearly impossible to contain transformative technologies like artificial intelligence.

DeepSeek isn’t the first Chinese company to adapt in the face of US sanctions. Tech giant Huawei rebounded to profitability after reshaping its business to counter similar restrictions. However, DeepSeek’s rapid ascent has sparked particular concern in Silicon Valley and Washington.

Venture capitalist Marc Andreessen likened it to a “Sputnik moment”—referring to the Soviet Union’s satellite launch that highlighted the technological gap with the US during the Cold War.

For years, many believed that US dominance in AI was secure, thanks to major players like OpenAI and Meta (formerly Facebook). Although China had pledged to lead in AI by 2030, its early offerings—including Baidu’s Ernie Bot—failed to impress, reinforcing the belief that Beijing’s strict tech regulations stifled real innovation.

This perception was supported by US policy under President Joe Biden, which focused on restricting Chinese access to the high-tech chips required for training advanced AI models.

However, DeepSeek’s achievements have challenged these assumptions.

“It’s upended long-standing beliefs about the scale of computing power and data processing needed for breakthrough AI,” said Samm Sacks, a senior fellow at Yale Law School’s Paul Tsai China Center.

She questioned whether “cutting-edge AI can now be developed with a fraction of the costs and computing power previously thought necessary.”

While American AI development strategies have traditionally emphasized large-scale investments, DeepSeek’s approach focused on cost-efficiency and performance optimization. This contrasts sharply with initiatives like former US President Donald Trump’s $500 billion Stargate project—a massive AI infrastructure program led by SoftBank and OpenAI.Remarkably, DeepSeek’s R1 chatbot was reportedly developed for just $5.6 million, raising questions about whether significant AI innovation requires hefty financial investments.

Some experts, however, urge caution. “DeepSeek V3’s training costs, while impressive, are consistent with historical trends of improved efficiency,” said Lennart Heim from the RAND Corporation.

He pointed out that “AI models have been getting cheaper to train over time—this isn’t entirely new.” Heim also noted that the full costs of infrastructure, research, and development might not be fully disclosed.

Nevertheless, Trump called DeepSeek’s success a “wake-up call” for Silicon Valley, urging tech leaders to remain “laser-focused on winning the AI race.”

Former US Representative Mark Kennedy argued that DeepSeek’s rise doesn’t necessarily weaken the case for future export controls. “Washington can respond by expanding AI chip restrictions and tightening oversight of technology transfers to Chinese firms,” he said.

Kennedy, now the director of the Wilson Center’s Wahba Institute for Strategic Competition, suggested that the US might also “increase domestic AI investments, strengthen global partnerships, and refine policies to maintain leadership without unintentionally driving more countries toward China’s AI ecosystem.”

Rebecca Arcesati from the Mercator Institute for China Studies (MERICS) concluded that “the real fear of losing ground to China may now trigger a renewed push for AI dominance in the US.”