Anthropic CEO doubts DeepSeek's efficiency claims, says US needs to limit China’s access to all AI chips

Dario Amodei, CEO of AI company Anthropic, has weighed in on the growing excitement around DeepSeek’s new Chinese AI model, questioning the true efficiency of the R1 model. While DeepSeek’s rise has stirred interest in the tech industry, Amodei believes the advantages it claims over US AI companies are exaggerated and short-lived. He argues that the costs of DeepSeek’s AI model are likely higher than reported, as it builds on prior research from US companies, as well as its own earlier V3 model.

Amodei’s comments come amid concerns over China’s potential to surpass the US in AI development, particularly if it gains access to millions of high-performance chips. He insists that the US must strengthen export controls on chips to China, as these controls are now even more critical than before to prevent China from taking a leading position in AI—and, by extension, on the global stage.

More from Tech

How ChatGPT is becoming everyone’s BFF and why that’s dangerous America ready for self-driving cars, but it has a legal problem

DeepSeek’s R1 model: Claims and reality

Amodei argues that while DeepSeek’s model is impressive, the real cost and sustainability of its advantages remain unclear. He suggests that American companies will soon be able to replicate these efficiencies, especially when paired with their superior access to more advanced chips and resources. According to Amodei, this could lead to the US regaining its competitive edge in AI.

However, he acknowledges that AI development requires substantial resources. The kind of advancements being discussed will likely take years, with the expectation that breakthroughs will occur around 2026-2027. Amodei stresses that it is the availability of chips, rather than just financial resources, that will ultimately determine whether China can keep pace with the US in this high-stakes technological race.

The stakes for global AI leadership

Amodei presents a scenario where, if China secures the necessary chips, the world could see the emergence of two superpowers—China and the US—both with advanced AI systems. This could lead to rapid scientific progress but also create an environment where military applications of AI become a critical factor. The US already has its share of partnerships with defence companies, with firms like OpenAI working alongside the US military on AI solutions.

While Amodei respects the contributions of Chinese researchers and acknowledges that AI should benefit all societies, his concern is about preventing China from gaining military dominance through AI. He stresses that the US and its allies must ensure they retain the upper hand in AI development, particularly by enforcing stringent export controls on chips.

Editor’s Picks

1

US Navy bars its sailors from using DeepSeek’s AI model 2

Microsoft investigating whether DeepSeek illegally used OpenAI's training data for their own model

Balancing technological power and ethical concerns

Amodei’s perspective, while diplomatic, underscores a central dilemma: Should the world be dominated by a US-led AI system or a China-led one? He suggests that export controls on chips are crucial in ensuring the US maintains its lead, even if temporarily. However, there’s also the pressing question of whether we can create a future where AI is not militarised or used to drive us to the brink of conflict.

The debate continues, with experts like Amodei calling for careful regulation and international cooperation to ensure that AI’s vast potential is harnessed for the benefit of all, rather than turning into a weapon of global dominance. Whether the US can maintain its lead or whether China will catch up remains to be seen, but the race is far from over.