NVIDIA reportedly buys Hugging Face for $12.9B; GLM-5.3-Flash and Qwen4 preview land

altryne · x · 2026-09-01

ThursdAI's Aug 27 episode rounds up the week in AI:

• The Information reports NVIDIA agreed to acquire Hugging Face for $12.9B, about 3x its 2023 valuation; the crew debated implications for open source for an hour.
• Z.AI's mystery OX Alpha declassified as GLM-5.3-Flash: 320B-A18B MoE, MIT license, served on Chinese chips.
• Alibaba previewed the Qwen4 architecture with Qwen3.8-Flash-Next; Google dropped Gemini Omni 1.1 Flash live, topping Arena's text-to-video leaderboard.
• Two killer interviews: TIME 100 honoree Andy Masley dismantled datacenter panic numbers one by one (including a 4,500x water error in Empire of AI); Kwindla Kramer unveiled PhoneLLM, an open-weights Nemotron post-train that jumped PhoneBench from 28% to 72% and runs voice agents at a quarter cent per minute.

Original post →

More from Venture

• Half of startup pitches focus on post-training open-weight models on proprietary data — tedgreenwald · 2026-09-01
• BoE chief warns inflated AI valuations and leverage could trigger next financial crisis — The Decoder · 2026-09-01
• NVLink adoption deepens investment in NVIDIA hardware — BenBajarin · 2026-09-01
• UK Creates Its First AI Unicorn Focused on Sovereign AI — agstrait · 2026-09-01
• Zhipu AI H1 Report: Revenue up 400%, API unit costs down 80% — 赛博禅心 · 2026-09-01
• Three months running Latentsync on 8GB locally: the real cost of "free" — cloudybrain07 · 2026-09-01

Venture channel →