What Would a ‘Sputnik Moment’ for US–China AI Competition Look Like?

On 4 October 1957, the Soviet Union launched Sputnik 1, the world’s first artificial satellite, into orbit. This launch sparked panic in the US on account of the perceived capabilities gap between Soviet and US space technologies. In 1958, President Dwight Eisenhower argued that if this gap was maintained, US leadership and national security would be threatened. Shortly afterwards, NASA was established. 

A similar alarmism has been present among many Western commentators following the recent release of cheap, high-performance AI models by the Chinese company DeepSeek. Commentaries on the subject have made wide-ranging claims. Some suggest that DeepSeek’s models are evidence of the US taking the wrong approach to AI development and of US export controls on semiconductor chips failing. The most alarmist comments have stated that this is a ‘Sputnik moment’ for AI development, with China ‘leapfrogging’ the US. 

Beyond the widespread media attention, there are few meaningful parallels between DeepSeek’s release and the Sputnik launch. The company’s models are impressive and may alter the market dynamics related to AI, but from a geopolitical perspective, there is little to suggest that China is outcompeting the US in advanced AI. 

The Economics of DeepSeek 

The development of advanced AI technologies like ChatGPT and Claude has been guided by the concept of scaling laws. These laws hold that the performance of machine learning models improves as the size of the model, the amount of data, and the computational resources used to train them increase. This has led companies to spend vast amounts of money, particularly on computational power, to build bigger models. The Stargate Project – a $500 billion investment in AI infrastructure for OpenAI announced earlier this month – is perhaps the starkest example of the bigger = better paradigm. 

DeepSeek took an alternative approach to simply building a bigger model, focusing instead on utilising innovative techniques to improve efficiencies. As a result, the training cost of its V3 model was only $5.6 million, a fraction of the estimated $78 million it cost OpenAI to train ChatGPT-4o. While the overall costs of producing the model, including hardware, labour and experimentation were significantly higher, the development nonetheless spooked the market. In a single day, US tech stocks suffered a loss of $1 trillion in market cap. 

There are two reasons for this market panic. First, efficiency improvements demonstrated by DeepSeek indicate that less computational power is needed to develop competitive models than the market had previously priced in. Second, DeepSeek’s performance is indicative of US AI companies facing fiercer competition from Chinese competitors than the market had previously priced in. 

Whether this sell-off was a blip or marks a turning point for AI development is currently unclear. Major US tech companies are transparent in their goal of developing ‘artificial general intelligence’ (AGI) and based on the current scaling paradigm, this will require efficiency gains and more computational power. If investors continue to support this aspiration because of the speculative benefits AGI offers, DeepSeek’s releases will not mark an inflection point. Conversely, DeepSeek could raise questions over the financial trade-offs of scaling, leading investors to focus more heavily on cheaper models and addressing ‘the last mile problem’ of integrating AI into real-world applications. 

The Geopolitics of DeepSeek

The geopolitical implications of DeepSeek are clearer cut. Many commentators have suggested that this is a ‘tipping point’ in US–China competition on account of the failure of US export controls, which have simply fuelled innovation within China. This is an optimistic view of the outlook for China’s AI industry. 

DeepSeek’s models were trained using Nvidia’s H800 chips that were sourced prior to the introduction of export controls targeting them in October 2023. These export controls may not have had time to fully impact China’s AI companies, and increasingly stringent export controls will be more difficult for Chinese companies to circumvent. Efficiency gains may be able to cover some of this computational power deficit, but DeepSeek’s founder has cited access to chips as the biggest challenge the company faces. 

In parallel with this, many of the techniques pioneered by DeepSeek are known and have been experimented with by Western companies. There remain technical challenges to implementing these efficiency gains, but given that DeepSeek and other leading Chinese AI models are open source, Western companies will likely be able to build on their developments with a computational power advantage. 

For China to compete on a level playing field with Western competitors, breakthroughs in domestic chip capacity will be required. This is something that has been recognised, with tens of billions of dollars set aside by national and local government to support domestic chipmaking capacities. Yet, there is significant ambiguity around whether these efforts will succeed and within what timeframe. Indeed, reporting restrictions on chip advancements are making it increasingly difficult to understand the progress China is making. 

It is always foolish to write China off – as sceptics have learnt with the breakthroughs made by DeepSeek – but it appears unlikely that China will surpass the US in advanced AI capabilities in the medium term, if the scaling paradigm holds. 

What Would a Sputnik Moment for AI Look Like? 

There are two likely avenues for a genuine Sputnik moment in US–China AI competition, defined in terms of US capabilities being surpassed. The first relates to the practical application of AI. 

DeepSeek’s focus on pushing the frontiers of development towards AGI is the exception rather than the rule in China’s AI industry. While the US has gone all-in on expensively pushing scaling at the frontier, Chinese companies have been competing in ‘the hundred model war’ to commercialise their products and cut costs. The focus in China on practical applications has led it to outperform the US in areas ranging from robotic installations to the construction of smart cities and the integration of generative AI into business processes. 

If Chinese companies continue developing relatively cheap and high-performing AI models which are better integrated into the real economy, then significant gaps in productivity and targeted innovation could emerge. Rather than a sudden realisation of a technological gap, as was the case with the launch of Sputnik 1, the compounding benefits of practical implementation in China could lead to a slow realisation by US policymakers and tech companies that they chose the wrong path for unlocking AI’s value. 

The second scenario, which would be a Sputnik moment in the more traditional sense, is Chinese companies making rapid developments in other types of advanced AI. While US companies and funding appear entirely focused on increasing the size and complexity of models as the best approach to develop more advanced AI, thinking in China is more diverse. Alternative approaches to developing advanced AI systems, like brain-inspired AI, are being explored by leading scientists and supported by national and municipal governments. 

The degree to which these alternative paths will prove fruitful for further advancing AI is unclear. That said, scaling has not always been the dominant paradigm for AI, and the methods being explored by Chinese scientists underpinned previous breakthroughs like Deepmind’s AlphaGo. Moreover, there are signs that certain aspects of the scaling paradigm may be plateauing, raising questions over the relationship between scale and performance. Placing all its eggs in one basket is thus a risky strategy for the US, which could see its lead in AI being undermined. 

© Huw Roberts, 2025, published by RUSI with permission of the author

The views expressed in this Commentary are the author’s, and do not represent those of RUSI or any other institution.

For terms of use, see Website Ts&Cs of Use.

Have an idea for a Commentary you’d like to write for us? Send a short pitch to commentaries@rusi.org and we’ll get back to you if it fits into our research interests. Full guidelines for contributors can be found here.