It won’t have escaped your notice that DeepSeek R1, the generative artificial intelligence chatbot released by a private Chinese company at the end of 2024, has attracted enormous attention. Initial headlines focused on its immediate impact on the financial markets as US tech stocks plummeted, with $589 billion wiped from Nvidia’s value on January 27 alone.
But DeepSeek could have broader implications. The chatbot matches the performance of its American competitors, such as ChatGPT, but comes free of charge. This raises the prospect that it might challenge America’s dominance in the field and allow other countries to develop their own cheap ‘sovereign’ AI systems that could bring the economic and societal benefits the technology has long promised but struggled to deliver. DeepSeek also intensifies the technological rivalry between China and the United States and has the potential to drag other countries into that confrontation.
Since the release of ChatGPT in November 2022, its creator OpenAI and other US tech giants have won huge investment and high market valuations largely because their AI products rely on the most expensive computer chips and a costly, data-intensive process to train the software. This has convinced successive US administrations to pursue a protectionist strategy.
Before leaving the White House, Joe Biden imposed export controls on computer chips to limit China and other nations catching up with the US. One of Donald Trump’s first actions as he began his second term was to announce Project Stargate, a $500 billion public-private initiative intended to expand domestic computing capacity.
Affordable innovation
DeepSeek has upended this status quo by innovating with more affordable lower-grade chips free of US export controls and developing a cheaper training process that requires less data and less computational power – both of which help cut costs. And while access to ChatGPT can cost as much as $200 a month, access to DeepSeek is free – though sometimes limited by popular demand.
‘Tech executives in the US have a huge financial interest in making us believe whatever they say,’ said Carissa Véliz, a researcher at the University of Oxford specializing in AI ethics. ‘What [DeepSeek] shows is that we can have models that don’t depend so much on data, and don’t depend energetically on so much computing power – and that changes the game.’
The arrival of DeepSeek’s R1 model also counters long-held beliefs about China’s ability to develop technology. ‘There is a very entrenched idea in the West, and particularly in the United States, that China cannot and will not innovate – it can only copy,’ said William Matthews, senior research fellow for China in the world at Chatham House. ‘What that means is that every time China produces something like DeepSeek, it comes across as a shock.’
Sidestepping censorship
Nevertheless, DeepSeek’s Chinese origins give rise to concerns over misinformation and security. When prompted, the chatbot declined to engage with questions about the diplomatic status of Taiwan or the 1989 Tiananmen Square protests. As for security concerns, it is worth noting the model came not from a government project, but a private company and, as such, the technology is open-source and can be freely downloaded and developed. This means any country or company can harness its top-of-the-line processing capabilities and free their model of censorship issues.
Governments are increasingly aware of AI’s ability to enhance efficiency and innovation. Sovereign AI put to public use could swiftly digest archived policy papers to unearth overlooked ideas, automate the process of drafting regulation and help in benefit assessment and prioritizing medical treatment based on likely outcomes.
Such hopes recall the early days of ChatGPT. Discussion focused optimistically on the potential for countries to develop sovereign AI systems, tailored to language, societal makeup and cultural differences and trained on proprietary data.
However, as tech companies realized the commercial potential of generative AI, the vision of capable AI systems, controlled by governments and paid for by the public sector, faded. OpenAI recently announced GPT Gov, a proprietary version of ChatGPT for the US government, which offers locally run AI models that meet US government data-security requirements. But most countries can’t afford a GPT Gov, nor have they the technical expertise to run their own systems. DeepSeek would appear to reduce these obstacles.
‘You could, for a reasonable investment, run a regional model on this,’ said Richard Whittle, an economist analysing AI at the University of Salford. ‘I suspect one of the things we’ll start seeing is investments in maintaining, developing and incorporating a powerful open-source model like this.’
Countries that will benefit the most and soonest are likely to be those already comfortable with Chinese technology used in their infrastructure. ‘This goes back to the point of value,’ said Matthews. ‘Third countries will have a choice between US and Chinese systems, and they’ll go for the cheaper alternative. You will start seeing China as the go-to partner for AI, particularly as the likely response from the US is going to be to double down on [chip] restrictions.’
Public-sector race
This suggests that the next era of AI development could move from private-sector competition to develop the fastest models to a more geopolitical, public-sector race – and a fork in the road.
On the one hand, the US is seeking protectionist policies to maintain its technological advantage in AI. On the other, apparently affordable open-source models from China are matching their US rivals’ performance with fewer restrictions for users. Countries considering how best to develop the benefits of a sovereign AI-supported government may effectively be making a difficult choice between China and the US.
It’s an issue that shaped discussions at the AI Action Summit held in Paris in February which brought together government and the private sector to agree on the future of AI. In a speech, JD Vance, the US vice president, made a pointed reference to countries making technology pacts with ‘authoritarian’ regimes, saying that ‘partnering with such regimes never pays off in the long term’.
Vance made clear he thought the world should accept American pre-eminence in the AI sector, adding that what he saw as the European Union’s overzealous attempts to regulate AI could stifle innovation. So far, no country has announced an intention to develop its own sovereign AI based on DeepSeek R1 – but as nations consider their options in the face of the Trump administration’s assertive position, some may turn to it.