This Week in AI — October 26–November 1, 2026
This was the week the AI industry opened its books. Anthropic published its first public financials, Big Tech reported record cloud growth alongside record spending, and investors made clear they now price the cost of the build-out as carefully as the revenue. The models kept shipping, but the cap table was the main story.
Key Takeaways
- Anthropic: The S-1 shows $16.3 billion in first-half revenue and a Nasdaq listing under "ANT", with nearly half of revenue flowing through AWS, Google Cloud and Microsoft marketplaces. If you buy Claude through a cloud marketplace, you're in the majority, and that channel is now a disclosed part of the business.
- Alphabet, Microsoft and Amazon: Cloud revenue grew 39% to 103% across the three, but Alphabet still fell 12.6% in a day on capex. Expect providers to push harder on committed-spend contracts as they try to show returns.
- Google: Gemini 4 Carbon leads on security benchmarks but is limited to vetted cyber defenders for now. Plan around Argon today and treat Carbon as a 2027 line item.
- Publishers and site owners: OpenAI, Google, Microsoft and Anthropic backed a single robots.txt token for opting out of AI training. Keep your per-crawler rules until each company actually switches it on.
Wall Street Reads the Fine Print
Anthropic Files Its S-1
Anthropic filed to list on the Nasdaq on October 26. The prospectus reports $16.3 billion in revenue for the six months to June, up about 1,300%, and an $80.7 billion net loss. Most of that loss is a $72.4 billion non-cash charge on convertible notes, and the operating loss was $693 million. Fortune reports a target of a Thanksgiving-week debut. The headline loss is accounting noise, so the number builders should watch is the marketplace share, which ties Anthropic's distribution closely to the clouds it competes with.
Alphabet Posts Its Worst Day Since 2004
Google Cloud revenue more than doubled to $30.7 billion, but Alphabet missed revenue and profit forecasts, posted roughly $51 billion in quarterly capex and negative free cash flow, and fell 12.6% on October 29. Microsoft did better: Azure grew 48% and capex roughly doubled to $39.4 billion, and the market accepted it. AWS grew 39% to nearly $46 billion, its fastest pace in almost five years. Demand for AI compute is not the question anymore. How each company pays for it is.
Meta Raises Equity, Nvidia Writes a Check
Meta laid out a plan to raise up to $95 billion for AI infrastructure and priced an initial offering at $625 a share. Nvidia committed $3 billion as part of a wider partnership on the Muse models. Issuing stock instead of borrowing more suggests Meta expects the credit market to stay tight. For anyone building on Muse, the Nvidia tie-up should mean better inference performance on Nvidia hardware first.
OpenAI Trims Its Round
OpenAI reportedly cut its new funding target from $30 billion to $20 billion at a valuation of about $1.25 trillion, with the WSJ reporting $11 billion secured and some investor interest "tepid." Coming in the same week as Anthropic's filing, it shows that a big valuation no longer guarantees easy capital. Keep a second provider set up for anything critical.
Models Built for Defense
Gemini 4 Carbon Goes to Cyber Defenders First
Google launched Gemini 4 Carbon on October 28 as its strongest model, scoring 88.2% on SEC-bench Pro for vulnerability discovery and 71.0% on CWE-bench for patching. Access starts with trusted defenders in Google's Fairwind program. Introductory pricing is $2/$10 per million input/output tokens, rising to $4/$20 later. A gated launch is now the standard way to release a security-capable model, so expect weeks to months between the announcement and getting an API key.
Mistral Large 4 Ships Open Weights
Mistral published the weights for Large 4, a 1-trillion-parameter multimodal model, on October 29 under a modified MIT license. It runs on vLLM and SGLang and is also available on Hugging Face, OpenRouter and Microsoft Foundry. The license is permissive except for one condition: companies offering it as a hosted service need a separate commercial license. Self-hosting a frontier-scale model is now a real option, as long as you're not reselling it.
Quick Hits
- AI training opt-out: OpenAI, Google, Microsoft and Anthropic will honor a shared
__usage_ai-trainingrobots.txt token, with rollouts ranging from "coming months" to early 2027. - Apple: Apple Intelligence is coming to China with Alibaba's Qwen powering selected features. Developer testing starts November 4.
- Google: Gemma 4 ran on a TPU aboard a prototype satellite, an early step for its orbital data center research.
- xAI: A federal judge let the NAACP's lawsuit over gas turbines at the Colossus data center proceed.
- Salesforce: Agreed to acquire Rune Technologies, which builds military supply-planning software.
- Microsoft: Published research tracking how North Korean threat actors use AI in their operations.
This Week in AI is published every Monday by the Lablab team.
