The contest for AI leadership between the United States and China is intensifying. China has released its latest frontier AI model, Kimi K3, as open-weight software, allowing users to access, modify, and build upon the model. Its release has prompted comparisons within the technology community to AI’s “Sputnik moment.”
The historical analogy is instructive. During the early Cold War, the United States was widely regarded as leading the space race. That perception shifted on 4 October 1957, when the Soviet Union launched Sputnik 1, the first artificial satellite to orbit Earth. The event prompted a reassessment of Soviet scientific and industrial capabilities, contributed to increased US investment in defence and research, and reshaped geopolitical expectations.
Kimi K3 has generated a comparable reassessment. Early evaluations suggest the model performs competitively with current US frontier models, including Claude and ChatGPT, across a range of reasoning, coding, and analytical tasks, while reportedly requiring fewer computational resources and lower development costs. It is also available as an open-weight model, allowing developers to use and adapt it without paying for access. This challenges an assumption underlying much of the current AI investment narrative; that ever-increasing compute expenditure and access to advanced chips are the primary determinants of AI leadership.
There is another notable development. It has been widely speculated that Kimi K3 was developed with assistance from earlier-generation US frontier models. Whether or not that is the case, recursive self-improvement; in which one generation of AI contributes to the development of the next, is becoming an increasingly important feature of AI progress.
From nuclear proliferation to capability proliferation
The Cold War analogy extends beyond the space race. The defining anxiety of the first Cold War was nuclear proliferation; the spread of weapons of mass destruction to an expanding set of state actors. The defining anxiety of this new contest is capability proliferation, with cyberattack capability as a primary area of concern. Advanced AI lowers the technical barrier to sophisticated cyberattacks, disinformation campaigns, and autonomous offensive operations. Sufficiently capable open-weight AI models, once released, are accessible to any actor with internet access.
What this means for AI valuations
China’s open-weight strategy raises questions about the commercial assumptions underpinning Western frontier AI companies. Current valuations of firms such as OpenAI (ChatGPT) and Anthropic (Claude) reflect expectations that frontier AI capabilities are scarce, defensible, and capable of generating sustained commercial returns.
If comparable or near-comparable models become freely available through open-weight releases, those assumptions may weaken. Many enterprise and government customers are likely to continue favouring Western providers because of regulatory requirements, security considerations, or data governance concerns. Nevertheless, each credible open-weight release expands the range of alternatives available to developers and businesses, potentially reducing the proportion of users willing to pay for proprietary frontier AI models. If that trend continues, valuations for some Western AI companies could come under pressure relative to current market expectations.
Actively managed funds, such as the Fintax Intl. Funds, that can steer clear of overvalued AI companies are, therefore, well placed to navigate the evolving investment landscape that AI innovation brings.