A house newsletter · daily · researched and written by the Kurzia AI
Saturday 7 November 2026 · Issue 6 · 5 min read
OpenAI at $1.25 trillion, Anthropic's cheaper Fable, and the AI spend stall
The story that matters most
TechCrunch · yesterday
OpenAI Series ?
Reported by Bloomberg, via TechCrunch, on Friday: OpenAI has raised $11 billion so far in a round valuing it at $1.25 trillion. The round has not closed; a source told Bloomberg it could total about $20 billion in the coming weeks, well short of the $30 billion target the same outlet reported in late September.
Hours earlier, Bloomberg reported that some buyers eyeing Anthropic's IPO see a valuation as low as $1.1 trillion, against the $2 trillion expected roughly a month ago. Both frontier labs repricing down at once — with post-midterm hostility to data centres adding regulatory nerves — is the number to carry into any growth-stage AI mark this quarter. Who gains: every startup buying tokens, whose input costs keep falling. Who loses: the late-stage holders who underwrote the summer's marks.
Rounds and exits
Applied Compute — Cloud for training and running AI models; stage not stated; $350 million at a $3.25 billion valuation; led by Elad Gil; reported by Crunchbase News.
Physicl — Synthetic 3D environments for training robots; seed; $10 million; led by Inovia Capital; reported by TechCrunch.
Vensum Power — Finnish deeptech trimming conversion losses as power reaches chips; Series A; €9 million; co-led by Lifeline Ventures, L&T Group and Eviny Ventures; reported by Tech.eu.
Pierre Computer Company — OpenAI is buying Pierre Computer, whose Code Storage git service stores agent-generated code for Lovable and Bolt; no price disclosed; deal awaits regulatory approval.
Clutch Security — In talks to be bought by Palo Alto Networks for $500 million plus milestone payouts, near 50 times its roughly $10 million ARR; reported by The Information, via TechCrunch.
What the AI labs and big platforms did
Claude Fable 5.1 — Anthropic's new frontier model, shipping with a twin — Mythos 5.1 — reserved for vetted cyber and life-science programs. Fable 5.1 costs an estimated 25% less than Fable 5 for typical workloads, up to about 45% less for agentic work, on cache reads priced at $0.25 per million tokens ($10 in, $50 out). Eligible enterprises get zero data retention. Cheaper long-horizon coding shrinks the cost base of every coding-agent startup — and of vendors selling against Claude Code. Anthropic
Inkling 2 Small — A 594-billion-parameter mixture-of-experts — only 17 billion parameters fire per query — scoring 31 on Artificial Analysis's intelligence index, level with Gemini 3.8 Flash and ahead of Mistral Large 4's 26. Weights due under Apache 2.0 next week, after safety testing, would make it the strongest open model outside China; until then it is closed, on Tinker only. That squeezes sellers of open-weight hosting and fine-tuning. A $5 billion round at $40 billion pre-money, led by Accel, is reported by Business Insider and unconfirmed. Thinking Machines Lab
Images 2.6 — OpenAI's updated image model keeps list prices unchanged but, per testers on Hacker News running its pricing calculator, token counts roughly quadruple at low and medium quality — a 1024x1024 low-quality image goes from $0.006 to $0.025. Several builders have rolled back to 2.5, calling the new version worse, and one leaderboard reportedly still ranks 2.5 higher. If you underwrite image-heavy products on list prices, re-run the calculator before the next board pack. openai.com · from the Hacker News discussion
Alexa text shopping — Selected US customers can now text Alexa to search, cart and buy on Amazon, with a waitlist and a broader rollout due in the coming weeks. The launch follows Amazon blocking Meta's Muse agent from its site in September: platforms with inventory are walling off their own agents. For consumer-agent startups, distribution — not capability — is the moat question. TechCrunch
For your portfolio
TechCrunch · yesterday
Ramp AI spend data
Ramp's corporate-card data, published Friday, shows token spend grew just 1.5% across August and September after 42% growth in July — even as usage volumes kept rising. OpenAI halved API prices on GPT-6 Sol and GPT-6 Luna in September, and Anthropic has said in a filing that price cuts on Fable 5.1 and Opus 5.5 cut about $600 million from revenue growth. October re-accelerated somewhat: spend up 8.9% from September 23 to October 23.
Ramp's lead economist Ara Kharazian notes almost all AI spend comes from venture-backed, high-growth tech companies, so a stutter in that cohort hits every lab at once. Tokens are a buyer's market — good for your portfolio's gross margins, awkward for anyone whose growth story is reselling inference. Ask each AI-backed founder: how much of this quarter's revenue growth is real usage, and how much is the labs' price cuts passing through as volume?
“Almost the entire spend on AI is from venture-backed, high-growth tech companies.”
SaaStr · yesterday
Vercel's agent playbook
Vercel's COO Jeanne DeWitt Grosser told SaaStr AI the company cut inbound lead qualification from 10 people to one using an agent costing under $5,000 a year; its in-house support agent Vertex handles 93% of cases for about $150,000 a year all-in; and automating 30 SDR workflows lifted quotas 30% in a quarter. Each agent was built by a three-person team: an engineer, a data scientist and the function's best performer.
Two harder signals sit under the numbers. Vercel will drop any tool without a developer surface — agents reach products through APIs, not by clicking screens — and one in three customer deployments now arrives from coding agents like Claude Code or Codex. Vendors selling GTM or support outcomes are priced against a weekend prototype. Ask each portfolio CEO: which function's best performer have you shadowed and documented yet — and would your product survive being called only by an agent?
“Our best AI SDR reviewed every single output to improve the flywheel. And then finally, she couldn't improve it anymore.”