2026-09-04
By Vadym · Generated with AI, curated by me
Today’s AI news is about the stack quietly re-sorting itself by economics — Nvidia is buying Hugging Face outright rather than out-competing it, Broadcom’s custom chip business tripled and the market still shrugged, and Anthropic and Google both shipped new models within a day of each other, competing as much on price as on capability. Underneath it all, a two-year-old startup just became Y Combinator’s fastest-ever unicorn.
Nvidia agreed on September 3 to buy the open-model platform Hugging Face in a deal worth about $13 billion — roughly $11.9 billion to shareholders plus up to $1 billion in retention equity for employees who join. CEO Jensen Huang said Hugging Face will “remain an open platform for the entire AI ecosystem,” and the deal is expected to close in the first half of 2027. It’s Nvidia’s second-largest acquisition on record, behind its $20 billion asset purchase from Groq.
Anthropic released Fable 5.1 broadly on September 1, alongside Mythos 5.1, restricted to vetted organizations for cybersecurity and biological-research use. Fable 5.1 adds a 1M-token context window, cuts prompt cache-read pricing by 75%, and Anthropic says coding agents built on it trigger about 60% fewer safety-safeguard interventions than on the previous Fable 5. Text output now carries invisible watermarking, with a detection API in private preview.
Google released Gemini 3.8 Flash this week, alongside a new cybersecurity-focused model aimed at government and enterprise customers. It's priced the same as its predecessor — 75 cents per million input tokens, $3.75 per million output — despite gains in coding and agentic-task performance. Gemini Enterprise also picked up pay-as-you-go pricing, token discounts up to 20%, and a zero-dollar base subscription tier.
AfterQuery, an AI training-data startup founded by two 22- and 23-year-olds, closed a round valuing it at $3.2 billion — a more than 10x jump from the $300 million valuation of its Series A just five months earlier. The company builds expert-verified reasoning data used to train frontier models. Its founders were in Y Combinator’s Winter 2025 batch, just 18 months ago.
Broadcom reported AI semiconductor revenue up 221% year-over-year to $16.7 billion on September 3, and CEO Hock Tan projected $230 billion in cumulative AI revenue by fiscal 2028, including $350 billion in AI semiconductor shipments to six hyperscaler customers over two years. The stock still fell roughly 6% after guidance came in below the market’s elevated expectations.
Step back and today is about the stack re-sorting itself by economics rather than capability. Nvidia is buying distribution, not compute. Anthropic and Google are competing on the price of running an agent for hours, not on who tops a leaderboard for a day. And the public market’s patience for AI infrastructure growth is visibly thinning even as the actual numbers — 221% chip revenue growth, a 10x startup re-rate in five months — keep getting bigger. The next phase of this industry may be less about what models can do and more about who controls the rails and can defend the margin.
— Boba
Curated by Vadym