Analysis suggests China’s open-weights AI strategy is outpacing US proprietary models
A recent analysis contends that China’s decision to release open-weights models is creating a superior global ecosystem, while US reliance on closed APIs and short-term profit motives risks a collapse in AI spending.
An analysis published on Hacker News, citing Robert Hart in The Verge, argues that China’s strategy of releasing open-weights AI models is surpassing the United States’ proprietary, closed-first approach. The article contends that American AI companies lack a technical moat beyond brand loyalty, as models are easily interchangeable via API. While US export controls on GPUs and data regulations limit Chinese companies' ability to provide global centralized services, China leverages open distribution to create a superior ecosystem. The piece notes that the performance gap between US frontier models and Chinese open models is closing, with an 80% probability that startups are using Chinese models. The author warns that the US focus on short-term profits and restrictive practices could lead to a collapse in AI spending, potentially harming the US economy.
The analysis posits that AI models themselves possess little inherent defensive value, with switching costs being superficial. In the engineering sector, where models are accessed via API, developers can swap vendors with minimal workflow disruption. Consequently, the true competitive advantage lies in enterprise services, such as contract deals and system connectivity, rather than the underlying technology. The author suggests that because technical loyalty is low, companies that lock customers into closed ecosystems are employing a losing strategy compared to those offering permissionless access.
Despite US government export controls on GPUs and strict data regulations that prevent sharing certain information with Chinese servers, Chinese firms have found a workaround. While these restrictions hinder their ability to offer global, centralized services similar to those provided by OpenAI or Anthropic, they have turned a compute disadvantage into a distribution advantage. By releasing models openly, Chinese entities commoditise the layer where American firms generate revenue, allowing their technology to be hosted, altered, and integrated into diverse use cases without restriction.
The performance disparity between US frontier models and Chinese open alternatives is reportedly shrinking. Highlighting this shift, a16z partner Martin Casado noted in The Economist that there is an 80% chance that any given startup is using Chinese models. The analysis argues that this open approach fosters a more effective global ecosystem, benefiting sectors ranging from manufacturing to scientific research, whereas the US model prioritises first-order profits over broader ecosystem benefits.
The author warns that the current US incentive structure, driven by short-term gains and government-imposed restrictions, could precipitate a severe downturn. With the US economy significantly driven by AI spending, a collapse in that investment could have damaging economic consequences. The piece concludes that while the US previously championed the open internet, its current locked-down business practices in AI are misaligned with the technology’s potential, urging a strategy that supports open technology for the public interest.
