Over the past 24 hours, money and infrastructure news dominated: financing, IPOs, compute centers, optical modules and memory price hikes all moved at once, while open-source model racing continued. Here is today’s briefing.

Illustration of AI news highlights
An illustration of the day in AI

1. Today’s Highlights

Meta eyes up to $95B equity raise:Funding AI infrastructure and superintelligence pushes; 2026 capex seen at $135B, with Nvidia, AMD and Broadcom each interested in up to $5B of shares(来源:Sina, Wallstreetcn)

Yunnan’s first county-level petaflop center:Compute infrastructure sinks further into prefecture and county level(来源:Seedao News)

Alibaba Cloud HPN wins CCF award:The high-performance network for AI data centers takes a first-class CCF Science & Technology Award(来源:Xinhua)

Enflame lists on the STAR Market:The DSA-architecture AI chip maker goes public, widening exit lanes for compute VCs(来源:Cailian Press)

800G optical modules drive Eoptolink growth:Q3 output grew QoQ; 800G shipments confirm cluster-build momentum(来源:Guandian)

Tang Jie’s paper reveals multimodal internals:The paper details a next-gen multimodal architecture; a new GLM looms(来源:36Kr)

Haidian releases 3T-scale open model:Per Sohu, the model keeps the open-source camp’s scale lead(来源:Sohu)

Survey says 79% of firms use under half their GPUs:Idle capacity contrasts with the buying boom(来源:Wallstreetcn)

Anthropic donates Agent Skills standard to Linux Foundation:Agent skill interfaces move to open governance(来源:SanYi Life)

Samsung Q3 profit soars 8x:The memory upcycle shows up in earnings; Korean memory giants report this week(来源:Securities Star)

Inference startup Qujing Tech raises over RMB 1.3B Series B:Capital keeps betting on inference-side “quality token” delivery(来源:Sina, Zero2IPO)

2. One-Line Takeaway

Capital is flowing up (equity raises, IPOs, Series B) while hardware settles down (county compute, 800G volumes, utilization doubts) — the bubble debate and real deployment run in parallel, and memory/optics earnings remind us the bottleneck this cycle isn’t only GPUs. (Compiled from public sources; details subject to original reports.)