A house newsletter · daily · researched and written by the Kurzia AI
Wednesday 11 November 2026 · Issue 10 · 4 min read
Anthropic's roadshow opens at $70–$78; Oura pays $625M for Counsel Health; Amazon cuts a model bill 98%
The story that matters most
Anthropic · yesterday
Anthropic IPO
Takeaway: $70.00 to $78.00
Anthropic launched its IPO roadshow Tuesday: 825 million Class A shares at $70.00 to $78.00 each, listing on Nasdaq as "ANT", per its own release. The Wall Street Journal, via The Verge, reports trading could start Thursday or Friday next week.
The offering could raise up to $64.4 billion at a valuation up to $1.4 trillion — Reuters had earlier reported expectations as high as $2 trillion — which finally gives OpenAI's $1.25 trillion private round a public reference point. Per techstrong.ai, from the filing: roughly $81 billion in annualized revenue as of October 31, against nearly $900 billion in multi-year compute commitments. Secondaries and LPs gain liquidity; funds still marking late-stage AI at private premia lose. What it changes: every AI mark in your book now has a daily public price to argue with.
Rounds and exits
Charta Health — AI chart review for billing accuracy; Series B; $55 million; led by Insight Partners; revenue up 10x in 12 months, per the company.
Nidus — Embodied intelligence for defense-factory robots; stage not stated; $30 million; backed by Kleiner Perkins and Costanoa; seven figures revenue in its first four months, per the company.
Sumary — AI accounting automation for European accounting firms; seed; $25 million; led by Singular; follows a $4.2 million pre-seed; reported by Tech.eu.
WhiteMatter — Security and governance for enterprise AI agents; Series A; $20 million; led by Blumberg Capital; 2.25 million agents under management, per the company; reported by SecurityWeek.
SPARETECH — Enterprise AI for industrial spare-parts management; Series B; $20 million; co-led by Blume Equity and Insight Partners; revenue up fivefold since early 2024, per the company.
General Translation — AI full-stack localization for developers; Series A; amount not disclosed; led by a16z; customers include Cursor, Ramp and Clickhouse.
Ancep — AI agents automating insurance placement for commercial brokers; stage not stated; $8.5 million; led by Creandum, with Kleiner Perkins; reported by Tech.eu.
Counsel Health — Bought by Oura for $625 million, weeks after Oura postponed its IPO; Counsel had raised $36 million, per PitchBook; reported by TechCrunch.
Proximal Ventures — Debut $27.5 million fund for pre-seed and seed AI-in-drug-development startups, checks of $500,000 to $1.5 million; raised by ex-SignalFire principal Sahir Raoof; reported by TechCrunch.
Crunchbase Unicorn Board — October added 34 unicorns worth over $96 billion — half under four years old — taking the board past 1,900 private companies, per Crunchbase News.
What the AI labs and big platforms did
Gemini Robotics 2.5 — Google DeepMind's newest vision-language-action model — the layer that turns what a robot sees into movement — scores 30–100% higher and runs 20–30% faster than the last version across 200+ tasks, per its own post. Apptronik's Apollo 2 humanoids kitted car parts on active Mercedes-Benz lines in Hungary and Germany at a 96% success rate, squeezing bespoke industrial-automation vendors and rewarding anyone building on DeepMind's stack. DeepMind also open-sourced ASIMOV-Jailbreak, a robotics safety benchmark: Gemini 4 Argon resists 99.5% of jailbreak prompts versus 82% for the previous model. Google DeepMind
For your portfolio
Strands Agents (AWS) · yesterday
Strands Decider
Takeaway: 98% less cost
Two Amazon engineers replaced Claude Sonnet 4.6 with Strands Decider — a small decision model that writes no text, only returns a probability for each option — for labeling internal tickets into 20 categories, and quality held at 98% less cost, per the AWS team's write-up.
The numbers: about $22,200 per million tickets on Sonnet versus about $500 on a SageMaker instance, and call time down from 4,215 ms to 897 ms. Any backed company routing, tagging or triaging through a frontier LLM is burning margin a 44x cheaper model could keep. The post's own caveat: it works only for a fixed catalog, high volume and short context — if the output needs explanation, an LLM stays. Ask each founder: which of our agent steps are really classification, and what happens to our gross margin when a customer's engineer reads this post?
ActivTrak · yesterday
ActivTrak Productivity Lab
Its report, built on behavioral data from 9,140 knowledge workers across 467 enterprises, found AI users' workdays span 8.4 to 9.1 hours versus 7.6 for non-users; heavy users also switch applications 15.2% more often and get shorter focus sessions (37.8 minutes versus 48.2 for light users).
It is the counterweight to the "AI lifted productivity 30%" board slide: measured behavior shows people working longer, not necessarily producing more per hour. Where the AI sits matters — embedded tools average 0.6 app switches per hour against up to 4.2 for browser-based ones. Before a founder quotes an AI productivity number, ask what is actually being measured — output, or hours — and whether the tool lives inside the workflow or in a tab the team keeps swiveling to.