When long-time tech investor Marc Andreessen called the unveiling of Chinese AI model DeepSeek a ‘Sputnik moment,’ he was reflecting the fear in the West that China may have caught up with, or even surpassed, the US in the race for artificial intelligence (AI) supremacy. This fear alone caused a $1TN correction in US tech stocks. AI chipmaker Nvidia saw a 17 per cent sell-off in one day. Energy stocks, including hydrogen and uranium, also took a beating as the market seemed to re-evaluate its assumptions about the high-energy intensity hitherto assumed to be essential for AI computations.

Any China watcher will warn you against hyperbolising either Chinese strength or weakness. This principle can be applied both to China’s economy and its tech sector. Over the last few years, the consensus has been that China is behind the US in the AI race and is unlikely to catch up. Proponents of this view cite China’s disadvantages in terms of hardware, high-level AI talent, government regulation, and even its capital markets.

Just six or so years ago, though, it was the opposing narrative that tended to dominate views of China’s economy. When the long-term investor and father of Chinese AI, Kai-Fu Lee, published his book AI Superpowers in 2018, the prevailing view was that China was beating the US in the race for AI. Observers argued that Chinese tech companies were exposed to greater amounts of consumer data, Chinese graduate students and engineers were hungry to compete, the Chinese government had a generally laissez-faire approach to the sector, and the Chinese tech ecosystem benefited from intense and rapid scalability.

The reality is often more nuanced than the views propounded by both Chinese tech bulls and bears. China has a history of going its own way when it comes to technological development. This history includes the creation of the country’s first successful nuclear bomb in 1964 and the development of tech giants including Alibaba, Tencent and Bytedance. Chinese scientists have a track record of bootstrapping and ‘making do’ with limited resources. Throughout the 1950s and 1960s, Chinese nuclear physicists toiled on, despite massive constraints, including the Soviet withdrawal of technical advisors in 1960. Likewise, Alibaba found a way to outdo Yahoo in the early days of China’s e-commerce market, with significantly less capital and personnel. More recently, the rise of short-video app TikTok has demonstrated that China can produce truly global, competitive products in the internet age.

Is DeepSeek’s R1 model truly a ‘Sputnik event’, as Andreessen and others have depicted it? When you go beyond the media hype and look at DeepSeek-R1’s technical capabilities, it does not appear to be as capable as US models, even though DeepSeek appears to have driven down training and inference costs significantly. Yes, the R1 reasoning model beat OpenAI’s o1 across the AIME 2024, MATH-500 and SWE-bench verified benchmarks; and yes, R1 is 27 times cheaper in tokens per dollar than o1. However, there is technical evidence to suggest that DeepSeek is underperforming in private tests across a range of real-world tasks beyond the benchmarks’ scope. DeepSeek servers were also down earlier this week as they failed to handle the stream of demand for their chatbot. Moreover, DeepSeek suffers from serious constraints of political censorship. When asked sensitive questions about the Tiananmen Square Protests of 1989, for instance, the chatbot refuses to answer.

Rather than believing the simple narrative that China may be beating the US in AI, the conclusions we should draw from the DeepSeek story are as follows. The Chinese tech ecosystem, which is largely privately-led and funded, is far more dynamic than people in the West give it credit for. Any trip to Shenzhen, Hangzhou or even Guangzhou will disabuse you of the illusion that China can’t innovate, or that it can’t survive the chokehold placed on it by ever-increasing US export controls. China has been able to find ‘good enough’ workarounds to develop its AI ecosystem, even with massive export controls on cutting-edge chips. There are also signs that they are advancing more quickly than expected in chip design and some aspects of chip manufacturing. DeepSeek itself was the product of the private sector and local open-source communities in China. The 39-year-old DeepSeek founder, Liang Wenfeng, started off using AI for stock trading before pivoting to building AI models using only local talents.

The US-China relationship will be dominated by continued tech conflict and the formation of a new arms race over AI. Further de-coupling and splintering of the software and hardware of the tech world between these two powers will continue apace. Historically, the US has increased its innovative capabilities in times of intense technological competition. The Manhattan Project emerged during the Second World War, precipitated by US fears that Nazi Germany would devise a fission bomb first. The Soviet success with the launching of the first satellite, Sputnik, in 1957 led to the creation of NASA and ultimate victory in the Space Race, culminating in the first successful moon landing in 1969.

It is likely that the Trump administration will place further restrictions on semiconductor export controls. The US will also use its vast military-industrial complex and the national security establishment to spur more investment and R&D in AI. In November 2024, the US-China Economic and Security Review Commission proposed a ‘Manhattan Project’-style strategy to achieve artificial general intelligence (AGI), i.e. the singularity at which super-intelligent AI systems surpass human capabilities across all cognitive domains. In his first week in office, President Trump dropped Joe Biden’s executive order on AI safety and announced project ‘Stargate’, a $500BN AI data centre project in partnership with OpenAI, Oracle and Softbank. US frontier labs are directing their heavy R&D spending towards capabilities breakthroughs, especially on the path to achieving AGI. The greatest value will be unlocked towards this end of the AI scale.

China, for its part, will also double down on AI and its significance. It is no coincidence that the humble Wenfeng was recently invited to a roundtable tech conference hosted by Chinese Premier Li Qiang. While the US will increase semiconductor export controls, China will use critical minerals restrictions, which could also hurt US advanced tech and manufacturing industries, including its chip and AI sectors. Whereas leading US research centres are strongly focused on increasing AI capabilities, Chinese companies are fixated on reducing training and inference costs. This will probably mean that China becomes more competitive in commercialising AI applications early on across a range of sectors, from consumer services and fintech to healthcare and surveillance. Both Beijing and Washington may force their local AI companies to support efforts to win the broader AI arms race through the development of dual-use technologies, with significant implications for the future of hybrid and conventional warfare.

DeepSeek has opened up the possibility of cheaper, more efficient modes of AI training. As English economist William Jevons observed in 1865, coal demand increased rather than fell with the creation of more efficient steam engines throughout much of the 19th century. A similar historical framework can be applied to the relationship between AI and energy demand. According to Jevons’s logic, demand for AI models and applications will increase as they become more efficient in computer-intensive training and inference. The net beneficiaries of this law of economics will be the big US tech giants including Meta, OpenAI and Microsoft. DeepSeek’s massive efficiency gains means that US frontier labs can look to a near future where they direct extra graphics processing units (GPU) capacity toward smarter models.

Historians looking back on this period may highlight a quite different event as the true Sputnik moment in the US-China tech race. Back in 2016, when Google’s DeepMind produced AlphaGo to beat a human Go-player, the Chinese government and tech community woke up to the fact that the US was at the forefront of the next technological and industrial revolution. Companies like DeepSeek are a product of Chinese fears of US AI supremacy, and we should expect many more DeepSeeks to emerge in the coming years as the AI arms race intensifies.