Zhipu closes ~$5B round for next-gen GLM self-training; Anthropic reportedly targets Nasdaq IPO in October

创业邦 · wechat · 2026-09-14

A roundup of AI industry news:

Zhipu raised $5B to fund next-gen GLM base models, a fully self-training pipeline (recursive self-improvement loop), and compute infrastructure including domestic chip adaptation and inference optimization.
Google/DeepMind is rumored to be close to recursive self-improvement (RSI), an unverified claim that clashes with recent calls from Anthropic, OpenAI, and Musk to slow frontier AI development.
Anthropic reportedly picked Nasdaq for an October IPO, with valuation expectations up to $2T; Q2 revenue exceeded $11.5B with its first profitable quarter.
Embodied-AI infighting: Mech-Mind's founder Shaо Tianlan accused "deal-assembling" robotics companies of fabricating revenue via data-collection centers and related-party transactions, naming Galaxy General; three other leading firms were reportedly reported by a peer, which all involved companies denied or declined to confirm.
Meta AI privacy flaw: the assistant could surface private questions and compile profiles including children from users' content; Meta has issued a fix.
Also: Nadella backs cautious superintelligence development and embedded evaluators; Anthropic, OpenAI, and Google discussed a joint AI standards body; IDC reports Q2 global server revenue hit $166.3B, up 52% YoY, driven by AI infrastructure.

More from Companies & People