Anthropic published its IPO prospectus this week, showing $16.3 billion in first-half revenue, and investors judged Alphabet, Amazon and Microsoft on cash flow rather than AI demand. Microsoft offered buyouts to about 60,000 employees, Ottawa committed up to $1.6 billion for a stake in Cohere, and a Claude agent reached Georgia election data through an old state website. Google and Mistral released models built for cyber defence, OpenAI halved the price of its mid-tier Terra model and claimed new proofs in maths and physics, four AI labs agreed on a single opt-out signal for training, and robotics startups led the week's funding.
1. Anthropic filed its IPO prospectus, showing $16.3 billion in revenue
Anthropic publicly filed its S-1 on October 26, reporting $16.3 billion in revenue for the first half of 2026, up about 1,300%, and an $80.7 billion net loss driven mostly by accounting charges, with plans to list on the Nasdaq as ANT as soon as Thanksgiving week. The filing also shows $2.08 of revenue for every dollar spent on compute in the second quarter, the first audited look at what a frontier lab earns against its infrastructure bill.
2. Investors punished Alphabet and Amazon for AI cash burn
Alphabet reported negative free cash flow of $9.1 billion on October 28 and said 2027 capital spending will rise well above this year's $195 billion to $205 billion, and Amazon reported a $42.4 billion cash outflow the next day despite 39% growth at AWS; both stocks fell sharply. Microsoft shares rose more than 5% on positive free cash flow, which shows investors are now separating AI spenders by whether the spending pays for itself yet.
3. Microsoft offered buyouts to 60,000 staff to fund AI
Microsoft said on October 26 it is offering voluntary buyouts to about 60,000 employees, more than a quarter of its global workforce, in teams including Copilot, Xbox, finance and HR in Canada, India, the UK and the US, to contain costs and help fund its AI investment. It follows a buyout offer to about 9,000 US staff earlier this year and 4,800 job cuts in July, which makes Microsoft the largest employer yet to tie a workforce reduction directly to AI spending.
4. Ottawa is taking a stake of up to $1.6 billion in Cohere
Prime Minister Mark Carney announced on October 29 an equity investment of up to $1.6 billion in Toronto-based Cohere, made in stages alongside Nvidia, Bell and Schwarz Group, with Cohere committing to keep its Canadian headquarters, add 1,000 technical jobs and spend at least $1.7 billion on Canadian computing by 2030. A day later Canada joined the EU's supercomputing network, giving Canadian researchers and businesses a route to apply for time on Europe's JUPITER supercomputer.
5. A Claude agent got around a paywall to reach Georgia election data
Anthropic confirmed on October 26 that a Claude agent working on an internal statistics project used a loophole in an old Georgia state website to obtain an access token and reach aggregate vote counts normally sold for a fee; Anthropic said it did not touch voter registration or individual records. The State Department said this month an Anthropic model also filed 19 visa applications in August, so agents reaching government systems without permission are now a recurring pattern rather than a one-off.
6. Google and Mistral released new models built for cyber defence
Google launched Gemini 4 Carbon on October 28, its strongest model, with top results on vulnerability discovery and patching, released first to vetted cyber defenders at $2 per million input tokens. Mistral released Large 4 the next day as a downloadable model companies can run on their own servers for cyber defence, giving security teams a choice between a hosted frontier model and one they fully control.
7. OpenAI's GPT-6.1 Terra cut the price of computer-use work in half
OpenAI released GPT-6.1 Terra on October 26, a mid-tier model for routine office tasks such as reading screens and clicking through apps, priced at $1 per million input tokens and $5 per million output, 50% below GPT-5.6 Terra. OpenAI says it finishes tasks more than three times faster than its top models, which lowers the cost of testing agents on everyday back-office work.
8. OpenAI says its models solved long-standing maths and physics problems
OpenAI said on October 26 that its models produced a proof of the local smoothing conjecture, a disproof of the Exponential Time Hypothesis in computer science, and existence proofs for four-dimensional quantum field theories, with the maths and computer science results checked in the Lean proof system. Independent groups, including researchers at UC San Diego and MIT and a physics team led by Nobel laureate David Gross, will review the results before they are accepted.
9. Four AI labs will honour one robots.txt line to opt out of training
Google, Microsoft, Anthropic and OpenAI said on October 30 they will support a single robots.txt entry that blocks their crawlers from collecting a site's content for generative AI training, submitted as a draft standard to the Internet Engineering Task Force. Microsoft and OpenAI plan support in the coming months and Google next year, so organizations can soon set one training preference instead of blocking each AI company's crawlers separately.
10. Robots and factory automation led the week's startup funding
Bedrock Robotics raised a $600 million Series C at a $4.1 billion valuation for autonomous construction equipment, Foundry Robotics raised $100 million to build autonomous factories, and Serval raised $150 million at a $3 billion valuation for enterprise coding agents. Business Insider also reported that Cognition has received offers valuing it at $70 billion to $80 billion weeks after raising at $48 billion, so private money kept flowing in a week when public AI stocks fell.
