China’s approach to artificial intelligence is diverging from Silicon Valley’s playbook: models are increasingly open, inexpensive to run, and designed for widespread application rather than a singular chase for speculative superintelligence. The recent launch and rapid uptake of Moonshot AI’s Kimi K3 underlines that shift and exposes a practical model for diffusing AI across an economy.
## Why Kimi K3 resonated
The release of Kimi K3 — a model whose capabilities rival the top systems from Anthropic and OpenAI, according to industry reaction — prompted intense attention. Demand briefly strained Moonshot’s capacity and the firm paused subscriptions as it scaled. Allegations that some Chinese labs trained their models on American systems have surfaced, but those claims remain unproven.
What matters beyond performance is Kimi’s distribution model: it is released as an open-weight model, meaning any company or developer can download the model, run it on local servers and adapt it for specific uses. That openness is now common among several Chinese providers, including offerings from Alibaba, DeepSeek, Z.ai and MiniMax. Firms typically monetize through adjacent services such as hosting, enterprise setup and customization rather than charging per-query fees.
## Chinese AI: open, low-cost and application-focused
The combination of accessible weights and lower operational costs positions Chinese AI as a platform for broad industrial adoption. Developers and businesses can build tailored applications — from e-commerce management to logistics optimization and targeted marketing — without incurring the high token fees associated with the closed frontier models of U.S. firms.
That accessibility flows from both strategy and constraint. Restrictions on sales of top-tier Nvidia chips to China were intended to slow the country’s progress, but they also pressured local labs to prioritize efficiency. Many Chinese models now deliver strong performance at a fraction of the compute cost of some American counterparts, making deployment on local infrastructure economically attractive.
## Closed U.S. models, data centers and environmental costs
By contrast, leading U.S. models remain largely closed and metered, with enterprises paying substantial token fees to access capabilities through APIs. To meet soaring demand, American firms are building large-scale data centers, an expansion that carries environmental trade-offs and has triggered local opposition in some regions. The capital intensity of pursuing artificial general intelligence (AGI) has also left several frontier firms unprofitable while investors bank on future returns if AGI is achieved.
Chinese companies, on the other hand, are generally concentrating on creating capable, efficient models that can be embedded into products and services. That pragmatic emphasis on near-term applications helps spread efficiency gains across a wider base of firms and workers.
## Public sentiment and regulatory guardrails
Public attitudes toward AI differ markedly between the two countries. A recent AI
