Anthropic leases lab space as demand for Claude 'skills' rises

AI agents are moving out of demo land and into actual company workflows. That sounds abstract, but the stakes are simple — once businesses rely on an agent to write code, review filings, or run finance workflows, they care less about chatbot flair and more about packaging, reliability, and raw compute. That is the gap Anthropic is trying to close right now. This week, it paired a big push around Claude “skills” and finance-ready agent templates with a new compute agreement meant to relieve a very visible capacity squeeze. ### What is a “skill,” really? A skill is basically a reusable bundle of instructions, scripts, and resources that gives Claude a repeatable way to do a specific job. Anthropic describes it as a folder with a `SKILL.md` file plus supporting materials, so instead of rebuilding an agent from scratch each time, teams can hand Claude a playbook for one task and let it load that playbook when needed. Think less “new model” and more “operating manual with tools attached.” (anthropic.com) ### Why is Anthropic leaning on this now? Because enterprises do not want a clever generalist — they want something that can survive production. Anthropic’s own developer materials now foreground skills alongside tool use, evaluations, prompt caching, and MCP, which is a tell. The company is steering builders toward a stack where the model is only one layer, and the durable value sits in tested workflows, governed data access, and reusable task logic. (anthropic.com) ### What changed this week? The clearest product move was Anthropic’s new finance-agent rollout on May 5. Those templates package three pieces together: skills, connectors, and subagents. In plain English, Anthropic is no longer just saying “build agents with Claude.” It is handing customers starter kits for specific jobs like KYC screening, valuation review, earnings analysis, and month-end close. That is a shift from platform to packaged workflow. (anthropic.com) ### Why does finance show up first? Because finance is where the pain is obvious and the budgets are real. Anthropic has been building toward this for months with Claude for Financial Services, Excel integrations, market-data connectors, and prebuilt finance skills. It also teamed up with PwC in February to push enterprise agents into regulated industries, which tells you this is not a side experiment — it is a go-to-market strategy aimed at firms that need audit trails, approvals, and tight controls. (anthropic.com) ### Where does the “leased lab space” angle fit? The bigger story is capacity. Anthropic said on May 6 that it signed a compute deal with SpaceX for all capacity at the Colossus 1 data center — more than 300 megawatts and over 220,000 Nvidia GPUs within the month. That came just one month after a Google-Broadcom agreement for multiple gigawatts of TPU capacity starting in 2027. When a model company is stacking deals that fast, it usually means demand is arriving faster than its normal infrastructure roadmap. (anthropic.com) The “renting rival lab capacity” framing is best understood as emergency scaling behavior inside that broader rush for compute. ### How hard is demand hitting? Hard enough that Anthropic says annualized run-rate revenue has passed $30 billion, up from about $9 billion at the end of 2025. It also said the number of business customers spending more than $1 million annually doubled from over 500 in February to more than 1,000 by early April. That is the kind of growth curve that can turn infrastructure planning into triage. (anthropic.com) ### What is the catch for customers? More packaged skills should make Claude easier to deploy, test, and govern. But heavy demand also means customers are more exposed to rate limits, capacity bottlenecks, and shifting economics around premium models. Anthropic has already had to raise usage limits and talk openly about compute expansion, which is good transparency — but it also confirms that demand pressure is not hypothetical. (anthropic.com) ### Bottom line? Anthropic is turning Claude from a model you call into a system you install into work. Skills are the packaging. Finance agents are the wedge. Compute is the constraint. And right now, the company is racing to solve all three at once. (anthropic.com)