Alibaba roadmap: Qwen explores RSI, new models up to 10T params, 20GW datacenters by 2032
teortaxesTex · x · 2026-09-22
Alibaba CEO Eddie Wu unveiled the company's next-phase AI roadmap at the Apsara Conference:
- Qwen: plans to train new models at 5–10 trillion parameters; exploring RSI (recursive self-improvement), where the model identifies its own weaknesses from real task feedback, designs experiments, constructs data, and continues training; focus on complex long-horizon tasks and unified multimodal understanding/generation
- Compute: T-Head launched the Zhenwu V900 chip at 3x M890 performance, clusters scalable to 500K cards; Alibaba Cloud targets 20GW+ of global datacenter capacity by 2032, saying long-term AI demand already exceeds supply
- Outlook: machine thinking is under 3% of human total today, could exceed humans 1000x; AI coding today is like "the lightbulb in 1882" — future tasks may be split into millions of subtasks run by millions of agents
More from Companies & People
- OpenAI reportedly rushing Codex Bot to rival Grok Bot, warns on recursive self-improvement — dotey · 2026-09-22
- Microsoft candidly describes what LLMs actually do in court filings — HildeKuehne · 2026-09-22
- Longtime AI Figure Starts MS/PhD at MIT, Advised by Joseph Jacobson, on AI Molecular Design — chaitjo · 2026-09-22
- Survey of 107 leaders: 60% see AI productivity gains but costs remain unpredictable — rohanpaul_ai · 2026-09-22
- Engineers push back as AI usage becomes a performance-review metric in big tech — datawithsuman · 2026-09-22
- Anthropic and Adaptyv launch protein design contest with $1M in Claude credits — Flomerboy · 2026-09-22