Anthropic shipped its strongest model and a rival to Excel before its IPO, companies spent five times more on AI than last year, and Nvidia showed what key-person risk looks like

The most capable AI model yet arrived, but for most daily work the cheaper one you already use is still the better choice. Around it, new data showed businesses spending far past their AI budgets, AI vendors moving into spreadsheets, ads and video, and the world's most valuable chip company caught with no plan for losing its CEO.

• Anthropic released Fable 5.2, now the top-ranked model, and Google's Gemini 4 Carbon came close on real work at a quarter of the cost
  My take: Fable 5.2 leads every major ranking, and on one test of real contractor jobs it finished 41 percent of projects end to end, double the previous best. Per-token prices fell 20 percent, but it is still expensive and uses a lot of tokens: at its highest setting a task costs about 2.6 times what Opus 5.5 costs. The honest early verdict from testers is that it is uneven. On large, open-ended projects it is a clear step up. For everyday tasks it is no better than Opus 5.5, and sometimes more stubborn. Google's Carbon, now on general release, found 57 of 105 planted bugs in one test against Fable's 58, for about 31 dollars against roughly 140. My advice: keep Opus 5.5 or a cheaper model as your default. Use Fable for one big job you have been putting off, start it on medium effort, and tell it at the start which actions need your approval. One cheap, useful first job: ask it to do a security review of any software you had built with AI this year, because early users report it finds serious problems that earlier models created.
• Ahead of a 1.3 to 1.5 trillion dollar IPO next week, Anthropic launched Claude Sheets, Whiteboards and Data to compete with Excel and Power BI
  My take: Anthropic is aiming to list before Thanksgiving, and its yearly revenue rate reached 81 billion dollars at the end of October, up 10 billion in one month. The valuation debate is mostly noise for a business owner. What matters more is the new suite: a spreadsheet built for people and agents to work in together, a planning board, and a tool that pulls live data from your CRM and database into charts or short narrated videos. Microsoft stock fell 2.2 percent on the news. Nobody needs to drop Excel, and Claude Sheets exports to it anyway. The bigger point is that the AI tools are becoming the place where work happens, not a helper on the side. Anthropic's own legal team built a review tool from its playbook that cut marketing reviews from three days to one, and none of them write code. If you run a small team, pick one recurring report or tracker this month and rebuild it inside your AI tool. You will learn fast whether your team should do more of its work there.
• A survey of 400 large US companies found AI spend per company rose 4.7 times this year, about 2.5 times more than they had budgeted
  My take: Companies planned for AI to cost per seat, like normal software. Instead, agents bill by the work they do, so spending grew with usage, not headcount. That is why firms like Apple are now cutting daily AI caps for engineers from 300 dollars to 200. Buyers also moved toward the expensive models, not away: the share of spend on top models went from 35 percent to 63 percent, and Anthropic now takes 55 percent of business spend against OpenAI's 31 and Google's 9. Open models are 4 percent of spend but 19 percent of usage, which tells you where the cheap, routine work is going. The useful news is that 55 percent of the new spend came from new budgets, not from cutting other software. For your business, budget AI the way you budget electricity or cloud hosting: a monthly amount per workflow, reviewed every month, with a cap. Then check which jobs truly need the top model. Most routine ones do not.
• OpenAI let advertisers put a brand-trained AI agent behind their ads, as ChatGPT reached a fifth of Google's search volume
  My take: Businesses can now register with OpenAI, track ad results in a dashboard, and set up a business agent that answers questions after someone clicks an ad. You give it instructions and product details, and it talks to the customer for you. OpenAI says ChatGPT now handles about 20 percent as many searches as Google, its ads earn at a rate of about a billion dollars a year, and users have not pushed back on them much. Google itself said AI answers are slowing its search revenue. The take here is that ChatGPT is becoming a place where people shop and compare, not just a place where they ask questions. If you sell to consumers, ask ChatGPT the questions your customers ask and see whether your business appears in the answers. And write down, in plain language, what a good sales assistant would need to know about your products and policies. You will need that document for any agent, whether you run it yourself or OpenAI runs it inside an ad.
• AI can now watch hours of footage, cut it, add animations and keep the video up to date, which makes video a normal tool for office work
  My take: The newest models can take a plain request like 'find the three moments where customers complained about onboarding in 20 hours of calls' and actually do it. They can also build animated charts, translate and dub, and record their own software demos. Most of this now runs inside the main AI tools, not separate video software. I would not start with marketing videos. Start inside the company, where the bar is lower and the payoff is quick. Three steps, in order. First, give the AI one recorded meeting with a real decision in it and ask for a three-minute highlight video. Second, give it last quarter's and this quarter's numbers and ask for a short animated chart that shows what changed and why. Third, once those work, set up a weekly update or training video that rebuilds itself from fresh data, so it does not go out of date in three months. Video used to cost too much for these jobs. It no longer does.
• Nvidia's CEO admitted an affair with a woman US officials believe is a Chinese spy, and the company has no succession plan
  My take: Jensen Huang runs a company that makes up about 7 percent of the S&P 500, and its stock fell 11 percent after the news. Congress wants him to testify, the board has hired outside lawyers, and reports say up to 7.5 billion dollars of Nvidia chips may have reached China through the spy's employer. For most businesses the chip politics are far away. The lesson about structure is not. Huang has around 60 direct reports and has long refused to plan for a successor, so the board now has no safe way to act. Most small and mid-sized companies have the same problem on a smaller scale: one person who holds the key customer relationships, the passwords or the know-how. Write down who would step in for your top two or three people for a month, and what they would need to know. If you also buy AI chips or cloud capacity, expect tighter export rules and more paperwork on where hardware ends up.