AI briefing: this week

This week at a glance

Five days into the week, the AI race has become a finance and construction race. Google is backstopping a roughly $19 billion lease so Anthropic gets 401 megawatts of Kentucky data-center capacity.

Anthropic's updated IPO paperwork showed about $33 billion in revenue over the first nine months of 2026 - roughly twelve times a year earlier - alongside much larger future computing commitments.

Fal, which rents out the computing power that runs other companies' AI, raised $1.3 billion at a doubled $16 billion valuation. Salesforce agreed to buy call-center voice-AI maker Cresta, reported at about $2.7 billion. And Anthropic signed the Associated Press as its first licensed news publisher.

On the product side, the open-source tier had a big week: Tencent launched Hy4, and Moonshot AI released the weights of Kimi K3.1 Pro - a 2.8-trillion-parameter multimodal model - for anyone to download.

ChatGPT learned to make animated GIFs, Google cut AI video prices 30% with Gemini Omni 1.2 Flash, Apple Intelligence finally got a China launch date with Alibaba's Qwen inside, and OpenAI said ChatGPT now handles a fifth as many searches as Google.

OpenAI released GPT-6 Cyber, a security model it classifies in its highest risk tier: built for vetted defenders to find and validate software vulnerabilities, and OpenAI says it has already been used to develop 948 Linux exploits and uncover 81 previously unknown Firefox flaws.

The week's most striking assessment came from outside researchers at METR, who spent 11 weeks inside Anthropic and found its model capabilities are now improving about twice as fast as a year earlier - largely, they believe, because AI is speeding up the research itself.

Claude moved into personal finance: US subscribers can now connect bank accounts through Plaid for read-only spending insights.

And the week's political turn: after midterm elections put data centers on the ballot, OpenAI promised towns it will pay its own power bills, publish site-level power and water usage, and dropped official NDAs. Google echoed the pledge a day later with a $17 billion Wyoming data center - its own fuel-cell power plant included - and a promise to pay for 100% of the power it uses.

Key stories of the week

1. Researchers at the outside testing group METR spent 11 weeks inside Anthropic and concluded its model capabilities are now improving about twice as fast as a year earlier, largely because AI is speeding up the research itself.
2. OpenAI released GPT-6 Cyber, a model that finds and validates software vulnerabilities for vetted defenders - and the first model OpenAI classifies in its highest risk tier for cyber capability.
3. Google is guaranteeing a roughly $19 billion data-center lease in Kentucky, so Anthropic gets 401 megawatts of computing capacity without buying the buildings itself.
4. Google plans to spend $17 billion building a data center in Wyoming that comes with its own power plant, and it promised to pay for all of the power the site uses rather than raising locals' electricity bills.
5. Anthropic's updated IPO paperwork showed about $33 billion in revenue over nine months - roughly twelve times a year earlier - alongside even larger future computing commitments.
6. After midterm elections turned data centers into a political issue, OpenAI promised towns it will pay its own power bills, publish site-level power and water usage, and stop requiring NDAs.
7. Claude can now connect to your bank account through Plaid to answer questions about your spending. The link is read-only - it cannot move money - and the feature is a US beta for paid subscribers.
8. Salesforce agreed to buy Cresta, a company whose AI runs call-center conversations, reportedly for about $2.7 billion.
9. Anthropic signed the Associated Press as its first licensed news publisher, paying to use AP reporting in its products.
10. Moonshot AI released the weights of Kimi K3.1 Pro, a 2.8-trillion-parameter model that handles text and images, free for anyone to run.

This week's briefings