US technological supremacy in artificial intelligence is being threatened by the rapid progress of Chinese open-weight models. According to expert Gary Marcus, the development of systems such as Moonshot.AI's Kimi K3, Z.ai's GLM 5.2, and Alibaba's Qwen is neutralizing the competitive advantage of the closed proprietary APIs used by Western laboratories.
What Happened
Gary Marcus stated that American closed models from OpenAI and Anthropic are losing their "moat." Chinese developments are reaching comparable quality while remaining available for local deployment. Instead of policies focused on export restrictions or bans on open source, Marcus suggests the US should implement a "CERN for AI" concept, turning fundamental developments into a public good for science and medicine.
Context
The traditional business model of leading US AI labs is built on exclusive access to powerful capabilities via cloud APIs. However, the increasing efficiency of open-weight models is shifting the industry focus toward local inference and weight optimization, turning access to basic AI capabilities from a scarce resource into a commodity.
Why It Matters for the Industry
For the industry, this signifies a paradigm shift: high-performance open models could disrupt current monetization models for proprietary systems and lead to the devaluation of the intellectual property of major laboratories. Startups will have to find value in applied tasks and specific data rather than simple access to models.
Why It Matters for Users
For end users and developers, this means reduced dependency on cloud providers and Western companies. The transition to open weights allows for the creation of autonomous, local, and private AI products using tools like vLLM or TensorRT-LLM for efficient model deployment on one's own hardware.
What Is Not Yet Known / Limitations
Expert opinions regarding the consequences of this shift are divided, ranging from predictions of large-scale market devaluation of intellectual property to the necessity of a radical change in government strategy.
Sources
Author
Look at AI, Editorial Team