← Daily AI news

Daily AI news, Tuesday, 10 November 2026

Daily AI / Tech / VC Briefing: Tuesday 10 November 2026

Anthropic logo and brand image
Image: Anthropic (anthropic.com)

Good morning, here's today's briefing. The lead: Anthropic briefly sent banks IPO terms showing a raise of nearly $70 billion, then pulled them back, as its roadshow starts today. Also today: Nvidia investigates whether Jensen Huang was targeted by a Chinese spy, Anthropic buys $3.4 billion of gas generators, Google adds two new TPUs aimed at agents, Automattic takes on Lovable, Nominal triples its valuation, Sakana AI eyes an IPO and a top Microsoft researcher walks out.

LEAD: Anthropic's IPO terms leak, then get retracted, on the eve of its roadshow

Background: Anthropic filed publicly for a Nasdaq listing under the ticker ANT on 26 October, with the price and share count left blank. Last week investors pushed back on a hoped-for $1.5 trillion valuation after a sharp sell-off in AI stocks, and Bloomberg says talks have since slipped to about $1.1 to $1.3 trillion.

What happened: Reuters saw emails showing an Anthropic representative sent banks a filing late on Monday with an offering of about 867 million shares at $70 to $80 each, which would raise $69.3 billion at the top of the range. Anthropic would sell about 843 million shares and existing holders about 23 million. A follow-up email said the pricing details were inaccurate and the filing would be updated later; Reuters couldn't tell what was wrong. The same draft showed $26 billion lined up in a private placement at the IPO price from Google, Nvidia, AMD, Broadcom, Micron, Microsoft and TSMC. The Wall Street Journal reports Anthropic is aiming to start trading next Thursday or Friday, with Goldman Sachs and Morgan Stanley leading, and the formal roadshow is expected to begin today. Reuters via StreetInsider, WSJ via StreetInsider, Investing.com, Yahoo Finance

Why it matters: Even as a draft, the numbers show the shape of the deal: one of the largest IPOs ever, backed by the chip and cloud giants that sell Anthropic its computing. The messy retraction adds a little uncertainty to a listing that will set the tone for every AI valuation, OpenAI's included.

Nvidia opens an investigation into whether Jensen Huang was targeted by a Chinese spy

What happened: The New York Times reported that US intelligence believed Alice Huang, a Chinese businesswoman (no relation) with whom Nvidia's chief executive had a close relationship, was recruited by China's top spy agency. She was a senior executive from March to November 2023 at Megaspeed, a Singapore data-centre firm now under Singapore government investigation over alleged evasion of US chip export rules. On Sunday Jensen Huang admitted being "unfaithful" to his wife and apologised. Nvidia's board launched an independent investigation, said it has "confidence in Jensen and his leadership team" and that it has no reason to believe national security was compromised. On Monday President Trump said the FBI is also investigating, and that Huang told him "I had no idea she was a spy." China called the allegations "entirely groundless." TechCrunch reports a race to sue Nvidia has begun, and Huang pulled out of the ElevenLabs Summit. BBC, AP, NPR, TechCrunch

Why it matters: Nvidia, worth about $4.5 trillion, has no public succession plan, and Huang has been its main negotiator with Washington and Beijing over chip sales to China. Any doubt about him lands on the company at the centre of the AI race, in the same week Anthropic tries to sell shares to public investors.

Anthropic buys $3.4 billion of gas generators to power its data centres

What happened: The Financial Times reports, and TechCrunch and CNBC relay, that Anthropic is paying $3.4 billion for Caterpillar gas generators able to produce up to 1.7 gigawatts. They were originally ordered by equipment supplier OnePWR, which couldn't finance the purchase; Anthropic went straight to Caterpillar and did not buy OnePWR. Waiting lists for gas turbines stretch into the early 2030s. Until now Anthropic has mostly rented computing, including $1.25 billion a month to SpaceX's xAI through May 2029, according to its filing. TechCrunch, CNBC, Tom's Hardware

Why it matters: Power, not chips, is becoming the bottleneck. A company that preferred renting is now buying its own power plant equipment to jump a six-year queue.

Google adds two TPUs built for agents: one for secrets, one for huge contexts

What happened: Google announced TPU 8tx and TPU 8i Max, making four new TPU systems in one year. TPU 8tx is its first with hardware-enforced confidential computing, keeping data and model weights protected even while being processed, paired with AMD EPYC hosts, for regulated businesses and governments. TPU 8i Max puts two 8i chips in one package for double the compute, on-chip memory and memory bandwidth, aimed at many agents working across a whole codebase or hours of video. Google says TPUs now have nearly 100 customers, and a new open-source TPU agent suite plugs into Claude Code, Cursor and Google Antigravity to help move code onto TPUs. No pricing or availability dates were given. Google, TechTarget

Why it matters: Google is selling its chips as a real alternative to Nvidia's, at a moment when Nvidia's leadership is under a cloud, and it's courting coding-agent users directly to make switching easier.

Competitor move: WordPress owner Automattic launches Spacefast to host vibe-coded apps

What happened: Automattic launched Spacefast, which gives apps built with AI agents the plumbing they need to go live: hosting, databases, sign-in, file uploads, email and analytics. Unlike Lovable or Emergent, it doesn't come with its own builder; you bring your agent, and it lists ChatGPT, Claude, Hermes, OpenClaw, Grok Bot, Muse, Pi, Cursor, OpenCode and Amp, or connect a Git repo. It's led by Batuhan İçöz, who started it in a company hackathon after friends kept sending him Vercel links, and it soft-launched in September. He pitches it to enterprises whose marketing teams are "churning out landing pages on services like Lovable." TechCrunch, PR Newswire

Why it matters: The fight is moving from who writes the app to who hosts it. If agents do the building, the valuable spot is where the result lives, and Automattic is betting on being neutral about which agent you use.

Funding: Nominal triples to a $3 billion valuation; Sakana AI eyes a 2027 IPO

What happened: Austin-based Nominal, which sells an AI platform for engineering and testing complex hardware such as aircraft and nuclear reactors, raised a $240 million Series C at a $3 billion valuation, led by Lightspeed with Insight Partners. That's triple the $1 billion it was valued at in March. It has 80 commercial and government customers, including Anduril, and won a $71.2 million US Air Force award in October. In Japan, Sakana AI has hired Goldman Sachs to advise on an IPO as early as autumn 2027 and is weighing listings in Tokyo and the US, which Nikkei says would be one of Japan's largest startup IPOs and its first generative AI lab to list. TechCrunch, SiliconANGLE, Nikkei Asia, FN News

Why it matters: Investors are still paying up for AI that touches the physical and defence world, and Anthropic's listing is opening a path for other AI labs to go public.

Key departure: John Langford leaves Microsoft and declares AI Frontiers "over"

What happened: John Langford, a founding leader of Microsoft Research's roughly 50-person AI Frontiers group, which built the AutoGen agent framework and the Fara computer-use agent, has left Microsoft. The group was folded into Microsoft AI in September. On his blog he described the new leadership's philosophy as "scaling, and aggressively not research" and wrote: "I am just not comfortable being at a company where the research I do has no path to real use." He called the episode "a bit of a train wreck" on X. India Today, hunch.net

Why it matters: The group behind one of the best-known open agent frameworks has been dissolved, and a senior researcher is free for any lab that still values agent research beyond scaling.

← Sunday, 8 November 2026 Wednesday, 11 November 2026 →

Get the weekly sift

The best new AI and agent releases, with one-line verdicts. Occasional picks, no spam.

Privacy policy: your email goes to our email provider, Resend. Unsubscribe any time.

News is editorial and never paid for. Sponsored content on ToolSift is always labelled.