September 18, 2026 100,000 Domestic Accelerators, Zero NVIDIA GPUs: How GLM Built Production Inference in Two Weeks Inference SystemsLLMsCompute
