A house newsletter · daily · researched and written by the Kurzia AI
Friday 13 November 2026 · Issue 12 · 5 min read
Huang resigns at Nvidia; Mirendil's $1B Series A; Gemini frontier prices double December 1; United blocks AI agents
The story that matters most
Tom's Hardware · yesterday
Jensen Huang is out at Nvidia, effective immediately
Takeaway: effective immediately
Nvidia's board confirmed Thursday that Jensen Huang resigned as president, CEO and director, effective immediately; CFO Colette Kress is interim chief while it searches for a successor. Nvidia has no succession plan behind Kress. For a fund: the supplier behind most portfolio compute is leaderless while export controls are politically radioactive — expect tighter enforcement, slower allocations, and harder looks at deals carrying a Nvidia strategic cheque (Nvidia joined the Agile Robots Series F the same day, below). Near term nobody gains; AMD and the custom-silicon shops gain optionality.
Rounds and exits
Mirendil — builds tools that let small teams run frontier AI research with agent swarms; Series A; $1B; led by Kleiner Perkins and Clearlake Capital, with Google among participants.
Agile Robots — builds humanoid and industrial robots; Series F; $500m; led by SoftBank, with Nvidia and Hitachi; reported by Tech.eu at an $8.5bn valuation.
Hatz AI — sells an AI platform for small businesses through managed service providers; Series A; $24m; led by Nadia Partners.
Nebuly — measures the business value of enterprise AI agents; Series A; $21m; led by Planven; reported by Tech.eu.
Biomatter — designs custom enzymes with AI for pharma and industrial customers; Series A; $18m; led by Astanor; reported by Tech.eu.
Norwest — Wells Fargo-backed firm closed its largest fund at $4bn, its eighteenth, to push deeper into physical AI, voice AI and agentic cybersecurity; reported by TechCrunch.
What the AI labs and big platforms did
Claude Whiteboards, Claude Sheets, Claude Data — three collaboration surfaces Anthropic shipped on November 10. Whiteboards and Sheets are in beta on all paid plans; Claude Data — the October dashboards product, renamed — is generally available with certified queries. Boards export to Miro and Lucid, sheets to .xlsx and Airtable. On Enterprise plans the first two are off by default, and Claude Data turns itself on by default on November 19, 2026. It presses on cheap whiteboard and BI startups; the bigger names become integration partners. Anthropic (claude.com)
Muse phone calls — Meta's consumer agent Muse can now ring US businesses for its users, but only for paid subscribers — plans start at $20 a month, per TechCrunch, with a $100 tier above it. With 10 million-plus users this pushes agents into phone trees and hold queues, flattening small consumer-calling startups like Instinct. The friction is on the other end: receptionists hang up on AI callers, and United's ban the same week shows hosts closing doors. Voice startups need sanctioned access, not human mimicry. TechCrunch
For your portfolio
Google (blog.google) · yesterday
Gemini 4 Hydrogen
Google shipped Gemini 4 Hydrogen on November 12: a fast execution model priced flat at $1.50 per million input and $7.50 per million output across its full one-million-token context window (the text a model can read at once), offered on every AI Studio tier including free, and now powering Google Search worldwide. It scores 51 on the Artificial Analysis index at its default setting versus 50 for Argon, and Google says it finishes tasks 2.5 times faster than Gemini 3.8 Flash.
Any product pinned to them for high-volume subagent or coding work sees its inference bill — the cost of running the model — double in under three weeks unless workloads move down to Hydrogen, or to Claude Haiku 5.5 and GPT-6 Luna, which Willison notes run $0.10/$0.50 per million tokens. The question for founders building on Gemini: which workloads are still pinned to Carbon or Argon, and what does the December 1 doubling do to 2027 gross margin if nothing moves?
CBS News · 2 days ago
United blocks personal AI agents from booking
United said on November 11 it will block third-party personal AI agents from booking flights and managing reservations — on its own digital platforms and through travel agencies and aggregators — and is rewriting its booking, ticketing and distribution policies. It cites credential sharing exposing data and risking unintended purchases or MileagePlus misuse, plus federal rules that require a human at check-in to confirm bags.
Concrete split: Southwest opened its inventory to ChatGPT, and Bank of America last week told customers not to share passwords with assistants. The companies hit are the on-behalf-of startups whose product logs into a host that never blessed them — the same wall Muse's new phone calls run into when receptionists hang up. Ask every founder with an agent acting inside someone else's product: if that host rewrote its terms tomorrow, is there a sanctioned API behind you, or just stored credentials?