Key facts
- China has developed an industrialized pipeline producing near-frontier AI models regularly.
- Chinese open-weight models now lead global usage charts.
- Chinese models offer efficiency, running tasks for cents compared to dollars on US systems.
- OpenAI cut Luna prices by approximately 80%, while DeepSeek raised V4 prices by 50% to 1,100%.
- Compute infrastructure remains a bottleneck for AI development.
- Companies providing AI infrastructure are the clearest beneficiaries of the current market.
The global artificial intelligence landscape is rapidly evolving, with China emerging as a significant player in the development and release of advanced AI models. On September 22, Xiaomi engineers broadcasted the training of their MiMo-V2.6 model, while Anthropic unveiled its Opus 5.5 model and OpenAI surprised the market with its GPT-6 Sol and Luna models.
This rapid pace of development is creating 'model fatigue' among tech executives and developers. Between June and August 2026, five Chinese firms launched six near-frontier AI models, including Moonshot's Kimi K3, Z.ai's GLM-5.2, DeepSeek's V4-Flash and V4 Pro, Alibaba's Qwen3.8-Max, and ByteDance's Seedance 2.5. This signifies an industrialized pipeline in China, capable of producing advanced models consistently.
Chinese open-weight models are now leading global usage charts, driven by their cost-efficiency. Tasks that cost dollars on leading US systems can be run for cents on Chinese models. This economic shift is occurring within a band, capped by open-weight competition and supported by scarce compute infrastructure. OpenAI recently cut Luna prices by approximately 80%, while DeepSeek increased its V4 prices by 50% to 1,100%, indicating a dynamic pricing environment.
The bottleneck for AI development remains compute infrastructure, as the unit costs of intelligence decrease while demand for accelerators, memory, space, and power remains high. Companies that provide this infrastructure are the primary beneficiaries, gaining from increased aggregate demand and usage-based billing for hosting open models. Model developers face challenges due to fast-rotating leadership, which implies thinner moats and makes premium valuations harder to justify. Enterprises and software deployers are expected to capture a larger share of the surplus.
Risks to this view include a sustained capability lead by closed-lab frontier models, potential compute restrictions in China, the hidden costs of hosting and securing open models, and a cooling of AI spending if frontier lab expenditures slow.