Microsoft's 60,000 buyouts, the AI-spending stall, and Anthropic's Thanksgiving IPO

🔥Today's one thing

🔍What is real, what is not

• Ramp's card-spend data — Companies' spending on AI grew just 1.5% across August and September, after 42% growth in July — usage kept rising but prices kept falling, so the bill did not. The vendors' endless-growth story is showing cracks just as Anthropic courts public investors, and it means a buyer has real leverage to keep demanding discounts. Anthropic's own filing says its recent price cuts cost it about $600 million in revenue growth. TechCrunch

  “Almost the entire spend on AI is from venture-backed, high-growth tech companies.”

• October US jobs report — Professional and business services added 70,000 jobs in October, far above the sector's 15,000 monthly average over the past year — one of its strongest months in a long while. The claim that AI is already wiping out office work did not show up in this month's data, though economists warn the skills employers want are changing fast and one report proves little. Part of the gain came from temporary staff, and the sector remains below its pre-pandemic trend. Business Insider

• Allstate and Manulife — Both reported concrete results on this week's earnings calls: Allstate's customer-service assistant, live since July, now handles 15% of chat volume, and Manulife is chasing $1 billion of enterprise AI value by 2027. Measurable agent results inside a traditional, regulated industry are worth more to your business case than another software-firm anecdote. Allstate also says 97% of its staff are now certified as AI-enabled. CIO Dive

📈Money and the market

• OpenAI's new round — Has completed roughly $11 billion in new funding at a $1.25 trillion valuation, people familiar tell Bloomberg, and the round could still swell to as much as $20 billion. The second-most-important private company in AI confirmed it can still raise at scale — and set the price the public market will judge Anthropic's listing against. The figures come from sources rather than the company, so the final size should be treated as still moving. Bloomberg (via Techmeme)

• Anthropic's IPO — is going ahead before the Thanksgiving holiday despite last week's market rout, with the company expecting to raise $80 billion to $100 billion at a valuation of $1.25 trillion to $1.75 trillion, the Financial Times reports. A listing within weeks brings public, audited numbers on the leading AI vendor's costs and customer churn — read them before signing any long AI contract. The report, published November 6, says the Claude maker is pressing on with the high-stakes float despite what the FT calls a brutal tech rout. Financial Times (via Techmeme)

• Claude Fable 5.1 — The new version of Anthropic's workhorse model is priced about 25% below its predecessor for typical use, and up to 45% lower for heavy automated work, with an enterprise option that keeps customer data entirely in the customer's own cloud. The price war is now official product strategy: if your AI contract was signed a quarter ago, the same work likely costs less today, and vendors are open to talking. List price is $10 per million units of input text, $50 per million of output. Anthropic

• Clutch Security — Palo Alto Networks is negotiating to buy the three-year-old startup for around $500 million, mostly cash plus later payments if targets are hit; nothing is signed, and neither side is commenting publicly. Cybersecurity for AI agents is turning into an acquisition race — CrowdStrike, Cisco and now Palo Alto have all paid up — a line your own security budget will eventually carry. The price would be near fifty times Clutch's yearly subscription sales, which are still under $10 million. TechCrunch

⚠️Risk the board will raise

• The DOJ's impersonation case — US prosecutors have charged three foreign nationals with pretending, using AI, to be executives and lawyers at major international law firms and persuading employees to move millions of dollars, in a scheme worth more than $129 million. This is the concrete version of the fraud question your board will ask: the senior voice instructing finance to wire money can now be manufactured, so payment-approval rules need to assume it. The indictment was reported by NBC News on November 6. NBC News (via Techmeme)

• Meta's terms reversal — Has indefinitely shelved a rule that would have forced businesses with a Facebook page to let Meta read their external websites for AI development, including via test logins past paywalls, after courts in Brazil and Norway blocked it and the UK opened a formal investigation. If you publish or sell through Facebook, the old terms stay in force for now; more broadly, the retreat shows courts and regulators can stop a platform's data grab faster than most expected. US senators of both parties had also demanded the withdrawal. Ars Technica

🌍The big AI stories today

• OpenAI — Its CFO said the midterm results didn't change its IPO plans, and Q3 revenue grew 70% with enterprise sales doubling — the election shock isn't denting demand. Business Insider

• Google — Is spending $17 billion on its first Wyoming data centre, fuelled by gas with carbon capture — the buildout continues despite voters' rejections. Google

• GPT-6 Cyber — OpenAI says it found 81 unknown Firefox flaws and built exploits for 948 Linux bugs with its gated security model; all figures are its own. BleepingComputer

💡One takeaway