US companies increasingly ditch American models for cheaper Chinese AI like GLM-5.2

The story captures a shift in enterprise buying psychology: with the price war compressing margins, US companies are testing and adopting Chinese open models — chiefly Z.ai's GLM-5.2, praised for coding and agentic tasks with a 1M-token context — because they deliver near-parity capability at a fraction of the cost. Community testers do flag painful latency as the tradeoff.

The investment community is validating the trend. CNBC reported Goldman Sachs picking favorite Chinese AI models and initiating coverage on Zhipu, while Invesco and Goldman analysts argued North Asian AI plays are now more attractive than their US counterparts. That's a notable reframing of the 'China is behind' narrative that dominated 2024-2025.

The development ties together several of the week's threads — DeepSeek's IPO and in-house chip push, Alibaba's Qwen maneuvering — into a coherent 'China is climbing the stack' theme. The caveats are real: data-residency and security concerns (amplified by the Alibaba/Claude Code China-detection allegations), export-control uncertainty, and reliability/latency gaps mean adoption is often for cost-sensitive, non-sensitive workloads rather than core systems. Vendor here is Z.ai/Zhipu, which falls outside the canonical vendor list.