Chinese AI models seize 61% global token share by Feb 2026 – JPMorgan
MiniMax, Kimi, and Zhipu AI dethrone US rivals on cost-efficiency.
New data from JPMorgan, released May 28, 2026, underscores a seismic shift in the global artificial intelligence market. According to the investment bank's analysis, by late February 2026, Chinese AI models collectively commanded 61% of total token consumption among the ten most-used models worldwide. The leading trio – MiniMax M2.5, Kimi K2.5, and Zhipu AI's GLM-5 – outperformed U.S.-based alternatives not just in usage share but in cost-efficiency, offering significantly lower compute costs per token. This rapid ascent suggests that Chinese AI developers have cracked the code for delivering high-quality inference at a fraction of the price, a factor that increasingly drives enterprise and developer adoption.
The JPMorgan report points to a structural advantage: Chinese models are purpose-built for large-scale, cost-sensitive deployments, leveraging optimized architectures and domestic supply chains. While OpenAI and Google still lead in benchmarks on some frontier tasks, the practical metric of token consumption reflects real-world deployment. If this trend continues, U.S. AI companies may need to slash prices or risk losing global market share. The data also hints at a broader trend: China's AI ecosystem is maturing from a fast-follower to a market leader in cost-optimized inference, a segment that could shape the next wave of AI adoption in business and consumer applications.
- Chinese models MiniMax M2.5, Kimi K2.5, and GLM-5 captured 61% of global token usage among top ten models by Feb 2026.
- Superior cost-efficiency is the primary driver, with significantly lower compute costs per token than U.S. rivals.
- JPMorgan data signals a structural shift: China increasingly dominates real-world AI deployment, while US models lead on frontier benchmarks.
Why It Matters
Cost-efficient Chinese models are reshaping the AI market – expect price wars and faster global adoption.