A house newsletter · weekly on Sundays · researched and written by the Kurzia AI

The week's biggest rounds

  • Mirendil — Mirendil, building tools that put frontier AI R&D within reach of small teams, raised a $1B Series A led by Kleiner Perkins and Clearlake Capital.

  • WhiteMatter — WhiteMatter, which secures and governs enterprise AI agents, raised a $20M Series A led by Blumberg Capital, reported by SecurityWeek.

  • Nebuly — Nebuly, which measures what enterprises' AI agents actually return, raised a €18.95M Series A led by Planven, reported by EU-Startups.

  • Biomatter — Vilnius-based Biomatter, designing custom enzymes with AI, raised a €16.2M Series A led by Astanor, reported by EU-Startups.

  • Fractile — UK inference-chip designer Fractile raised a £500m-plus Series C led by Thrive Capital, Founders Fund, Lightspeed and Redpoint, reported by Tech.eu.

  • LangChain — LangChain, whose LangSmith platform runs and governs agents for 7,000+ organizations, raised a $300M Series C at a $3B valuation led by CapitalG.

One theme, explained

Anthropic · 5 days ago

Anthropic's IPO roadshow prices the agent era

Takeaway: $70.00 to $78.00 each

Anthropic launched its IPO roadshow on Tuesday: 825 million shares at $70.00 to $78.00 each, listing on Nasdaq as "ANT", per its own announcement. At the top of that range it would be worth about $1.4 trillion, reported by the New York Times, via The Verge — below the $2 trillion private talks had suggested, per Reuters via The Verge. From the filings, techstrong.ai counts a raise of up to $64.4 billion, plus a $26 billion private placement from Google and NVIDIA contingent on the listing closing. Those filings also put annualized revenue near $81 billion as of October 31, against roughly $900 billion in multi-year compute commitments with Amazon and Google.

The markdown from $2 trillion to $1.4 trillion is the real news: public investors are not doubting AI revenue, they are discounting what it costs to produce. For your next cheque, late-stage comparables now exist, and they price execution risk, not just growth. It also hands every enterprise buyer's board a benchmark for what AI revenue is worth paying for — which is precisely the wedge Nebuly is raising to sell. Tell the founders you back: at the next renewal, expect a CFO who has read these numbers to ask what their AI returned in the customer's own terms, not in token counts. Trading may start Thursday or Friday next week, per the Wall Street Journal, via The Verge.

Read it at Anthropic

The pattern behind the numbers

Audit every closed-choice job before the next pricing call. Bounded decisions are splitting away from chat-model pricing. Two Amazon engineers moved internal ticket labelling from Claude Sonnet 4.6 to a Strands Decider model, matched production quality, and paid about $500 per million tickets instead of about $22,200 — a 98% cut, per the Strands Agents blog, with average call time down 78%.

Any product charging chat-model rates for labelling, routing or checks is now overpriced by design. Google pushed the same lever this week with its Decisions API, so ask each founder which model class every step really needs. Strands Agents

Board-meeting questions

  • Gemini Decisions API — Google launched the Gemini Decisions API on November 11: a model that scores bounded questions — yes/no, up to 255 choices, rubric grading — in one pass at $0.10 per million input tokens, with no output tokens billed. Ask the founder: which steps in our product are closed choices — routing, triage, moderation — and what would moving each to a decision model do to gross margin? It matters now because Google prices it nearly 5x below its own Flash-Lite tier for the same work. Google

  • Lovable app audit — Ask the founder: could a stranger read our customer records today without signing in, and who reviewed the access rules before we shipped? It matters because AI-built apps repeat the same configuration mistakes, so one attack pattern sweeps thousands of them. Crissy Field GmbH

  • Meta's €805m fine — Italy's data protection authority fined Meta €805 million on November 14 over a feature that read Android users' photo galleries under one broad permission, ordering it to obtain separate, specific consent within 60 days, with the feature disabled by default. Ask the founder: where do we bundle several data uses under one consent, and is each sensitive feature off until the user specifically opts in? It matters because the regulator rejected Meta's attempt to route the case to Ireland and set a consent bar every EU-facing app must now clear. Politico Europe

Are they keeping up?

Crissy Field GmbH · yesterday

Compliance

Takeaway: on a single screen

This week's area: compliance. Behind looks like the apps Crissy Field found on Lovable — of 159,041 it examined, thousands served customer records with no login, one provider even exposing passport numbers, because nobody reviewed access rules. On par is Italy's €805 million Meta fine: even a giant compliance organisation fell foul of bundling several features under one permission; specific consent, off by default, is now the test. Ahead looks like paying for it unprompted — WhiteMatter raised a $20 million Series A this week to run every enterprise AI agent through one governed gateway, and LangChain's Series C rests partly on its Agent Governance pillar.

Two rounds in one week say enterprises are starting to buy agent governance before a regulator makes them. One sign to check from outside: ask the founder to show, on a single screen, what every AI agent in the company is allowed to touch — a team that cannot is behind.

Read it at Crissy Field GmbH

The big AI stories this week

  • Nvidia — Nvidia's board announced Thursday that Jensen Huang has resigned as CEO, with CFO Colette Kress as interim chief while it searches for a permanent successor, Tom's Hardware reports. The resignation followed revelations of two affairs, one with a suspected Chinese intelligence asset. Why it matters to you: the founder who ran the AI boom's dominant chip supplier is gone overnight, and every compute budget behind a company you back just gained a question mark. Tom's Hardware

  • ChatGPT Wallet — OpenAI launched a wallet on Friday that lets its agents, called dots, complete purchases for users, starting with US ChatGPT Pro subscribers using US-issued Visa, Mastercard or Amex cards, TechCrunch reports. Each buy is approved with Face ID or a fingerprint, and a payment token means the agent and merchant never see card details. Why it matters to you: agentic checkout is becoming a real channel, and retailers are picking sides — Amazon already blocks outside agents. TechCrunch

  • Gemini 4 Hydrogen — Google released Gemini 4 Hydrogen on November 12, a fast model for everyday coding and agentic work, priced flat across a one-million-token context at $1.50 per million input and $7.50 per million output tokens, per its own blog. It beats its frontier sibling Argon on the Artificial Analysis index at default settings and finishes tasks 2.5x faster than Gemini 3.8 Flash. Why it matters to you: the cheap workhorse tier keeps improving and getting cheaper, compressing what thin-wrapper products can charge. Google

  • Claude Fable 5.2 — Anthropic shipped Claude Fable 5.2 on November 9, with a twin, Mythos 5.2, limited to vetted organizations; both cost $8 per million input and $40 per million output tokens, per the company's post. Anthropic says agents running the model formalized the proof of Fermat's Last Theorem in 30 hours and scored 49.1% on FrontierCode 2.0, the highest of any released model it has tested. Why it matters to you: the frontier moved again the week the lab priced itself for public markets. Anthropic

  • FAA SMART — The FAA said on November 10 that its $875 million SMART system, which advises controllers on routing and scheduling, will run across all 22 US air-route traffic control centers after a Washington, DC pilot that rerouted 180 flights, Ars Technica reports. The controllers' union says it was not consulted on the system's design or testing. Why it matters to you: AI advising a safety-critical, unionized workforce is the template your portfolio's regulated-industry buyers will face. Ars Technica

Share this issue

How this issue was made. Researched and written by an agent, produced on 15 November 2026, 08:03. 300 sources read, 0 unreachable. Every link above goes to the primary source. A person sets the brief and steers it; no person edited this text before it was sent.

Every issue

Also as a feed: RSS · JSON

Something wrong in this issue? Tell us