China AI labs slash costs as Moonshot, Z.AI and DeepSeek surge

This piece is one of the clearest signals yet that the cost structure of frontier AI is shifting from a U.S. monopoly on cutting‑edge models to a genuinely multipolar market. Moonshot’s Kimi K3 and Z.AI’s GLM‑5.2 are not just good enough for domestic Chinese users; they are already being quietly adopted by U.S. startups, Fortune 500 companies and developers who are staring down runaway API bills from OpenAI and Anthropic. When a model like DeepSeek V4 Pro can undercut a leading U.S. model by more than 50x on token price while staying within striking distance on quality, the old assumption that “whoever spends the most on GPUs wins” starts to look shaky. ([fortune.com](https://fortune.com/2026/07/26/china-moonshot-deepseek-zai-kimi-challenging-us-ai-cost/))

For the race to AGI, this matters in two dimensions. First, it lowers the marginal cost of experimentation, letting more teams and more countries run serious agentic systems, not just prototypes. Second, it undermines the leverage U.S. policymakers thought they had via export controls. If China can keep pushing model efficiency on domestic hardware and distribute permissively licensed weights, we end up with a world where powerful systems are cheap, forkable and politically hard to constrain. That accelerates diffusion of frontier‑like capabilities even if no single lab makes a dramatic capabilities leap.