The writer is a geopolitics analyst specialising in China
Stripping away the hype around DeepSeek, the Chinese AI start-up’s achievement can sound almost mundane. It leveraged known engineering tricks to create products nearly as good as top US models — just much cheaper, though not as cheap as some of the jaw-dropping estimates of its costs. DeepSeek is not a trailblazer yet. It is, however, still a game-changer.
Anthropic chief executive Dario Amodei wrote that DeepSeek did not achieve “a unique breakthrough” but rather reached “an expected point on an ongoing cost reduction curve”. What was unexpected, he said, “is that the company that was first to demonstrate the expected cost reductions was Chinese”.
It should not have been unexpected. The optimisation of processes to drive cost efficiency gains has long been a Chinese manufacturing superpower. This ability has propelled Chinese companies to domestic dominance and then overseas expansion in industry after industry: shipbuilding, solar panels, batteries, drones, electric vehicles, industrial robotics and pharmaceuticals. Now, it is coming for Silicon Valley and investors betting on US exceptionalism should take note. The US can no longer take comfort believing that it rules software even while China dominates hardware production.
The rise of Chinese industrial champions usually follows a pattern: a new technology — often originating in the US — emerges. The technology defuses to China, sometimes through state-orchestrated tech transfers and sometimes through “reverse engineering” combined with incremental innovation. The Chinese government classifies the industry as strategic. Legions of businesses sprout up to seek subsidies. Overcapacity leads to brutal price wars, forcing firms to relentlessly cut costs and compress margins. Finally, the survivors emerge, ruthlessly efficient and adaptable.
DeepSeek notably deviates from this pattern. It has emerged from the periphery of the state-driven, techno-industrial system and was self-funded by a former hedge fund manager. But China’s strategic necessity to survive computing power scarcity imposed by US export controls has spurred DeepSeek to develop greater engineering efficiency. As a start-up, it risks further US tech sanctions and market-access restrictions by distrustful governments before it matures. The US, South Korea, Italy and Australia are among countries that have partially banned or restricted DeepSeek. But it could end up like TikTok, which has been banned or partially banned in a few countries, but is certainly used at scale.
DeepSeek has plenty of opportunities to optimise usage through application in China’s techno-industrial cluster. Dozens of Chinese automakers, semiconductor design firms and financial companies have integrated DeepSeek into their products or processes. Such tie-ups are likely to generate mutual benefits of product enhancements and process efficiencies, even if some applications probably amount to mere trend-following.
American AI companies have been the pioneers and they are still pushing the frontier. Amodei is right that the cost efficiency of a pioneer and that of a follower are not comparable. Pathfinding is inherently inefficient. When the route is unknown, one cannot optimise the journey. But it becomes a systemic problem for the US when its companies pioneer new technologies in areas such as electric vehicles but lose the scaling game to their Chinese rivals.
Scale advantage compounds itself. Volume compensates for low margins. A broad user base fuels rapid product refinement. Frequent iterations drive continuous process optimisation. As a result, Chinese national champions often outcompete incumbents and establish deep market moats. Few challengers have the stomach to commit massive capital outlays only to fight painful price wars for wafer-thin profits. Dominance in one sector also often accelerates breakthroughs in adjacent technologies. Chinese battery makers’ success in enhancing product reliability and affordability gave rise to Chinese mass-market EVs and drones.
DeepSeek should worry Silicon Valley. US tech companies have long shown an ability to scale and reap the advantages — see Alphabet, Meta, Apple and Microsoft. But American manufacturers of solar panels, batteries and drones would warn that pioneering newer technologies is a shortlived triumph if US companies cannot learn to make their products quickly, reliably and cheaply. Failure at scaling risks falling behind Chinese competition.
