Today in AI business — November 10, 2026

Five items that matter to a small business, with sources. News moves fast and vendor announcements are vendor claims — check the linked source before acting on any of it.

1. Anthropic's IPO details circulated to banks, then retracted hours later

Details of Anthropic's blockbuster IPO — including a plan to raise up to $69.3 billion — were briefly circulated to banks on Monday, November 9, then retracted late the same day, according to emails seen by Reuters. The Information reported that Anthropic's outside counsel in Canada sent banks an updated prospectus and then withdrew it, saying some details were incorrect. The roadshow is still expected to formally launch later today, Tuesday, with pricing expected next week. The retracted figure sits below the $80–100 billion the Financial Times reported on November 6 that Anthropic aims to raise, at a $1.25–1.75 trillion valuation — and since the filing itself was withdrawn as incorrect, treat the $69.3 billion as unconfirmed, not as the plan.

Why it matters: Anthropic's listing is the reference point the whole AI sector is waiting on, and a botched filing days before listing week is the wrong kind of signal — even if the timetable holds. The number to watch now is the price range when the roadshow opens, not the figure in a withdrawn document.

Reuters: Anthropic aims to raise up to $69.3 billion in blockbuster IPO →

2. Anthropic releases Claude Fable 5.2 and Mythos 5.2 — faster and 20% cheaper (vendor announcement)

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on Monday, November 9 — the same underlying model in two configurations. Fable 5.2 is the generally available version, in paid Claude plans, Claude Code, the Claude Platform and the major clouds; Mythos 5.2 drops some cybersecurity and biology safeguards and is limited to organisations approved through Anthropic's Cyber and Life Sciences Verification Programs. Both generate outputs up to 30% faster than the 5.1 versions, and the standard API price falls 20%, to $8 per million input tokens and $40 per million output tokens, from $10 and $50. In Anthropic's own pre-release tests, the model helped solve two open problems in black-hole physics and formalised the proof of Fermat's Last Theorem in 30 hours, and in a separate experiment 1,000 agents built a working operating system from scratch over eight days. Independent evaluator Artificial Analysis ranked Fable 5.2 first on its Intelligence Index v5, with a score of 65.

Why it matters: A cheaper, faster flagship days before the IPO is Anthropic showing the business works — and it puts price pressure on OpenAI and Google at the top of the market. For a small business buying API access, the practical effect is simple: the most capable Claude yet costs a fifth less per token than its predecessor.

Reuters: Anthropic unveils new Fable, Mythos models as competition heats up ahead of IPO →

3. Google unveils two more TPU systems, TPU 8tx and TPU 8i Max (vendor announcement)

Google announced two more eighth-generation TPU systems on Monday, November 9 — TPU 8tx, built for secure and confidential inference on sensitive data, and TPU 8i Max, which puts two TPU 8i chips in one package to double compute, on-chip memory and memory bandwidth for massive context windows. It is the first time Google has introduced four TPU systems in a single year, after the TPU 8t for training and TPU 8i for real-time reasoning arrived in April. Google published no prices and no availability date for either system — Bloomberg reported that Google's Amin Vahdat said they arrive next year — and the performance figures are Google's own, not independent benchmarks.

Why it matters: Google is building a full chip stack for the agentic era: locked-down inference for regulated industries, and huge context windows for agents working across whole codebases. Every credible alternative to Nvidia hardware loosens the grip on AI infrastructure prices — but with no prices or dates, this is a roadmap, not a product you can buy yet.

Google: Our newest TPUs: 8tx and 8i Max, built for the agentic era →

4. Fireworks AI widens its October breach disclosure — billing data, datasets, unfinished models

AI infrastructure provider Fireworks AI said on Monday, November 9, that its October security incident reached further than first reported: attackers accessed billing information — addresses, invoice and payment histories, account balances and credit usage — plus secrets, datasets and incomplete model weights for a limited set of accounts. Payment card numbers were not affected; Fireworks says it does not store them. The company first disclosed unauthorised use of internal credentials on October 6, then a second intrusion on October 23–24, when a third-party cloud provider hosting its research machines was itself breached. Fireworks says it detected that activity the same day and has found no further unauthorised access since October 25, and has brought in Palo Alto Networks' Unit 42 for forensics. The company attributes the attacks to a financially motivated group selling stolen model access — a scheme it calls LLMjacking. Forbes reported that two customers have substantially moved their business away.

Why it matters: If you run models or store data on an inference platform, your intellectual property and billing data sit behind the vendor's security. This breach shows the blast radius when that fails — datasets and model weights, not just logins — and the two customers walking away show the commercial cost lands on the vendor too.

The Neuron: Fireworks AI was hacked — how criminals profit from stolen AI →

5. Moonshot AI looks to Europe for Blackwell rentals as Nvidia's reported cap bites

Chinese AI lab Moonshot AI has been holding talks with European cloud operators about leasing Nvidia Blackwell capacity — clusters of 8,000 or more GPUs — CNBC reported on November 5, with one source saying the lab offered to pay 30–50% more upfront than a non-Chinese customer. Renting chips housed abroad remains a legal grey area: US export controls block the physical sale of advanced chips to China, but not remote access to them through overseas data centres. The talks follow a Financial Times report on November 3 that Nvidia verbally told cloud partners to direct at least 51% of their services to non-Chinese customers on new contracts; Nvidia has neither confirmed nor denied it. The scarcity shows in prices: a B300 server rack that sells for about $600,000 internationally reportedly fetched $2.8 million in China in October.

Why it matters: The rental workaround is the pressure valve for China's AI labs, and if it narrows, the compute gap widens — while the same scarce GPUs get bid up for everyone else. For a small business, that is the other side of AI prices: the cost of the hardware underneath every model you rent.

Tech Times: Moonshot AI sought Nvidia Blackwell rentals in Europe →

No affiliate partnerships yet — links above go to the news sources and tool sites directly. When affiliate links are added, they will be clearly labeled, and editorial picks stay independent of who pays.

For what the AI companies and their leaders are saying on social media — claims and opinions, not verified news — see The Social Wire →

Browse the full directory →   Back home →

This day has a permanent page: November 10, 2026. Every day is kept — see the News archive. The News page always shows the latest day.

← Latest day · News archive · Previous day: November 9, 2026