Late October 2026: Anthropic opens its books

The month in AI is now twice a month: a short look back at what happened in AI and what it means for organizations putting it to work.

In early October, companies started putting AI on a budget. By the end of the month, investors were doing the same. Chip stocks fell, research firms questioned whether AI spending will pay for itself, and the largest tech companies were judged on their cash. Into that market came Anthropic's IPO filing, the first full look at the finances of a frontier AI lab. Here is what we think is worth your attention, followed by the rest of the two weeks' notable news.

1. Anthropic opens its books

Until Oct 26, companies building on Claude had to piece together their supplier's business from press reports and the occasional figure Anthropic chose to share. That day Anthropic made its IPO filing public, and for the first time the numbers were on the table.

The business is large and growing fast. Anthropic brought in $16.3 billion of revenue in the first half of 2026 and made an operating profit in the second quarter. The founders keep control, with a majority of the vote on most matters. We went through the details in our preliminary read.

It also rests on enormous fixed costs. Anthropic has committed to roughly $637 billion of computing capacity over the coming years, and paying for it takes borrowing on a large scale. Banks and Blackstone lined up $63 billion to buy Broadcom chips and lease them to Anthropic. Four days after the filing, the Financial Times reported that Blackstone, which had expected Anthropic to earn an investment-grade credit rating, now expects one just below it, in the top tier of "junk."

Meanwhile, most of that revenue comes from customers who pay as they go and are not locked in. Replit, until recently one of Anthropic's five largest customers, showed what that means on Oct 17, when it made OpenAI the default in its coding agent, though Claude remains available.

Why it matters. For the first time, companies that build on a frontier model will be able to read their supplier's finances every quarter: what it charges, what it pays for compute and how much it owes. That is useful in a price negotiation. It is also a reminder that a supplier with heavy fixed costs will look for ways to raise what each customer pays, so it is worth knowing how easily your own work could move to another model.

2. Investors want the payoff

Anthropic's filing landed in a market that had started to doubt. For two years investors rewarded companies for spending on AI. In late October they began asking what that spending had returned, and the answers were thin.

Chip stocks moved first, falling eight sessions in a row through Oct 16, their longest losing streak since 2008. The research that followed pointed the same way. Gartner's chief forecaster estimated that AI would need to earn more than twice what Gartner expects by 2030 to pay back what is being built. A Gartner survey found that only 13% of CIOs say their company has seen significant value from AI. Apollo's chief economist put it most bluntly: outside the largest tech companies, AI has shown "zero financial payoff" so far.

Then the largest tech companies reported earnings, and investors made the test plain: they wanted to see cash as well as growth. Microsoft grew its cloud business 48% and its shares rose. Alphabet doubled its cloud revenue, but its profit and ad sales fell short of forecasts, it burned cash in the quarter and it plans to spend even more next year. Its shares fell more than 10%. Amazon grew its cloud business at the fastest pace in nearly five years, yet it has spent $42.4 billion more cash than it brought in over the past year, and its shares fell almost 7%.

The builders are also turning to outside money. Meta plans to sell up to $97.5 billion of new stock, and SpaceX took on $40 billion of debt to buy AI chips. Lenders noticed. On Oct 23 the extra interest investors demand to hold corporate bonds reached its highest level in 17 months, partly on worries about AI borrowing.

Why it matters. The market is now asking every AI buyer's question, only louder: what did the spending return? Companies that can answer with numbers, in revenue per employee, cycle times or costs avoided, will find their budgets easier to defend in 2027. Those that cannot will face the same discount Alphabet did, from their own boards.

3. Large employers reorganize around AI

Inside large companies, AI is starting to change who does the work, and employers are taking two very different routes.

Some are using it to shrink. Barclays said it would cut about 5,000 jobs this year, mostly in support and operations, and credited AI and other technology with letting fewer people do the same work. Microsoft offered buyouts to about 60,000 employees, roughly a quarter of its staff. Microsoft has not named AI as the reason, though Reuters reported the aim was to contain costs while it pays for AI.

Others are putting AI in everyone's hands. Uber is taking ChatGPT Work from a small pilot to more than 30,000 employees, and Canva revealed an in-house AI research lab of more than 160 people.

Why it matters. Cutting roles and rolling out tools are two different bets, and neither delivers value by itself. Gartner's 13% figure suggests that most companies handing out AI have not yet changed how the work gets done, and few of those cutting staff have shown that the remaining work holds up. The companies that see a return will be the ones that redesign specific processes and measure them, whichever route they take.

4. Buying companies to rebuild them with AI goes mainstream, and regulators take notice

One of the most direct bets on AI's payoff is to buy an ordinary business and run it with AI. Until recently that was the work of a few specialist firms. In late October it reached a $15.3 billion deal.

On Oct 29 ADT, the home security company, agreed to be bought by Thrive Holdings. Thrive Holdings was set up by the venture firm Thrive Capital to buy established businesses and rebuild them with AI, with OpenAI as a partner supplying models and engineers who work inside the companies. Smaller versions of the same idea appeared across industries. Formation Bio launched a $600 million fund to buy drug programs and develop them on its AI platform, and Re:Build raised $2.1 billion to build manufacturing businesses. Anthropic itself bought Mithril Medicine, a startup whose founder had used Claude to coordinate drug development across 11 outside labs.

Regulators, meanwhile, showed how they view a long string of acquisitions. On Oct 19 the Federal Trade Commission and 12 states sued to stop PlayPower, a playground equipment maker owned by Platinum Equity, from buying its rival BCI Burke. The complaint pointed to three decades of PlayPower buying competitors, rather than to this one deal. On Oct 27 PlayPower dropped the purchase.

Why it matters. Buying businesses in fragmented markets and rebuilding them with AI is now backed by some of the largest investors in technology. The playground case shows that regulators will judge a run of acquisitions by its combined effect, even when each deal is small. Anyone building a roll-up, AI-driven or not, should expect the whole string of deals to be reviewed, not just the latest. We looked at how these buyers operate in the Bending Spoons playbook.

5. AI finds the holes, for attackers and defenders

Finding a serious flaw in software used to take rare skill and a lot of time. Late October made clear that AI is changing that for both sides.

On Oct 21 PHP and Node.js, two pieces of software that much of the web runs on, shipped fixes for flaws that could let an attacker take over a server. AI helped find flaws in both. The security firm Calif found one in PHP using an OpenAI model, and OpenAI describes two months of working with the maintainers on the fix before it was announced. In Node.js, OpenAI's own security agent reported one of the flaws.

Attackers are using the same tools. Two days earlier Reuters reported that a single hacker had used AI agents to break into more than 100 company websites, including Deere's, for about $50 per break-in.

The AI labs are finding the same problem inside their own walls. On Oct 20 OpenAI disclosed that two of its internal coding agents had weakened the system meant to watch them, and that one then denied doing it.

Companies are also deciding which outside AI agents to let in. Walmart opened a gateway for shopping agents, with spending limits each customer sets, while Amazon keeps blocking them. Danske Bank launched an official way for business customers to connect their banking data to tools such as Claude and ChatGPT. We wrote about why banks need that kind of front door in Agents at the bank's door.

Why it matters. Finding a hole in software now takes a model and a modest budget, on either side. The practical response is unglamorous: patch quickly, ask whoever runs your websites and applications whether this month's fixes are in, and decide which outside agents may reach your systems before they show up unannounced.

Notable AI news, Oct 16 to 31

• Oct 16. Chip stocks fell for an eighth straight session, their longest losing streak since 2008.
• Oct 16. Politico reported Anthropic's draft federal AI safety bill, which safety advocates criticized as too weak.
• Oct 17. Replit made OpenAI the default in its coding agent, keeping Claude as an option.
• Oct 18. Anthropic committed 5% of its shares to a public benefit fund.
• Oct 19. Gartner found only 13% of CIOs report significant value from AI, and fewer than a third have rules for stopping a pilot.
• Oct 19. Gartner's chief forecaster warned that $3.5 trillion of AI infrastructure spending is at risk of write-downs by 2030.
• Oct 19. Reuters reported one hacker using AI agents to break into more than 100 company websites.
• Oct 19. The FTC and 12 states sued to block PlayPower's purchase of BCI Burke, citing its history of buying rivals; the deal was abandoned on Oct 27.
• Oct 19. Tenstorrent, the AI chip company led by Jim Keller, raised $1.5 billion at a $6.2 billion valuation.
• Oct 19 to 20. US federal civilian agencies were ordered to remove DeepSeek's products.
• Oct 20. OpenAI disclosed that two internal agents weakened their own safety monitoring.
• Oct 20. Salesforce agreed to buy Braze for $5.66 billion.
• Oct 21. Anthropic opened a research preview of Claude Duet, an agent that works in the background across connected accounts and checks in before acting.
• Oct 21. PHP and Node.js shipped security fixes for flaws found partly with AI.
• Oct 21. Walmart opened a gateway for third-party shopping agents, with spending caps and approvals.
• Oct 21. Danske Bank launched an official connection from business banking data to AI tools.
• Oct 21. Legora, the legal AI company, raised $550 million at an $8.8 billion valuation.
• Oct 22. Barclays said it will cut about 5,000 jobs this year.
• Oct 22. DeepSeek reportedly raised about $15 billion, with CATL and Tencent as the main backers.
• Oct 22. Axios reported AI agents operating in customers' bank accounts; a follow-up found no rogue activity affecting customer funds.
• Oct 23. Global corporate bond spreads hit a 17-month high.
• Oct 23. OpenAI resumed frontier training after a security overhaul.
• Oct 23. Uber is rolling out ChatGPT Work to more than 30,000 employees.
• Oct 24. The UK set out about £600 million for AI security.
• Oct 24. Apollo's Torsten Slok argued AI has shown no payoff outside big tech.
• Oct 25. Elon Musk said SpaceX closed $40 billion of debt to buy AI chips.
• Oct 26. Anthropic filed publicly to go public. Our read is here.
• Oct 26. Banks and Blackstone lined up $63 billion to finance Broadcom chips for Anthropic.
• Oct 26. Microsoft offered buyouts to about 60,000 employees.
• Oct 26. OpenAI's models came to Google Cloud, payable from existing cloud commitments.
• Oct 27. Meta raised its planned stock sale to up to $97.5 billion.
• Oct 27. OpenAI was reported to be planning to buy Google's AI chips outright.
• Oct 27. JPMorgan was picked to lead the US retail part of Anthropic's IPO.
• Oct 27. Re:Build raised $2.1 billion to build manufacturing businesses.
• Oct 28. Microsoft reported Azure growth of 48%.
• Oct 28. Foundry Robotics raised $100 million to build an autonomous factory in Emeryville, California.
• Oct 29. Alphabet reported Google Cloud revenue of $30.7 billion, double a year earlier, and shares fell more than 10% after hours.
• Oct 29. Amazon reported AWS growth of 39% and a $42.4 billion free cash outflow over twelve months.
• Oct 29. Thrive Holdings agreed to buy ADT in a $15.3 billion deal.
• Oct 29. Mistral released the weights of Mistral Large 4, its trillion-parameter model, delivering on its end-of-October promise.
• Oct 29. Cohere announced the first close of a round of up to $5 billion, led by Germany's Schwarz Group, with the Canadian government among the investors.
• Oct 30. Blackstone was reported to expect a high junk rating for Anthropic after its IPO.