Alibaba Roadmap: 20GW Global Compute by 2032, Zhenwu Chips 3x Faster
kevinsxu · x · 2026-09-22
Per Kyle Chan, Alibaba laid out an ambitious roadmap: 20GW of global compute by 2032, next-gen Zhenwu AI chips 3x more powerful with mass production starting Q1 next year, and Qwen 4 in training with future models reaching 5-10 trillion parameters. The poster cautions not to sleep on the Zhenwu line while everyone watches Ascend.
Related event: Alibaba Roadmap: 20GW Compute by 2032, Homegrown Chips Run Qwen Inference(2 posts)→
More from Infra
- TRL async GRPO adds LoRA sync via storage bucket and proxy, cutting 500-step training to 53 min — SergioPaniego · 2026-09-22
- GGUF models can now run directly in Hugging Face transformers with ggml Metal kernels — pcuenq · 2026-09-22
- Tech men joke about footwear after Allbirds pivots into a data center company — chrisalbon · 2026-09-22
- Modular's LLM Inference Handbook adds a GPU architecture primer for kernel work — blaizedsouza · 2026-09-22
- Quantized GLM MoE model (W4A16 AWQ) trends on Hugging Face — AikidoSec · 2026-09-22
- OpenRouter silently ignores effort params — benchmark via native API endpoints, not assumptions — PawelHuryn · 2026-09-22