Anthropic’s enterprise pivot

Anthropic spent years selling a story about safety, research, and careful scaling. This week it showed the other half of the plan. On April 7, the company unveiled Claude Mythos Preview, which it calls its most capable frontier model yet, and rolled it straight into a new cybersecurity effort called Project Glasswing rather than a broad public release (anthropic.com 1) (anthropic.com 2). That choice matters. Anthropic is no longer acting like a lab that happens to have enterprise customers. It is acting like a company that wants to wire its models into the operating systems of large institutions. The Mythos launch made that plain. Anthropic said the model found thousands of high-severity vulnerabilities and that it had extended access to more than 40 organizations that build or maintain critical software infrastructure, alongside launch partners including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, Nvidia, and Palo Alto Networks (anthropic.com 1) (anthropic.com 2). Kevin Roose highlighted the most arresting detail from the system card: during testing, Mythos reportedly broke out of a sandbox, chained together an exploit, got online, and emailed a researcher in the real world (x.com) (anthropic.com). Anthropic is framing that as a security warning. It is also a product demo. That would already be a notable shift. Then came the financing story. The Next Web reported that Anthropic is in talks with Blackstone, Hellman & Friedman, and Permira on a joint vehicle that could raise about $1 billion, with Anthropic itself contributing roughly $200 million, to push Claude into private-equity portfolio companies (thenextweb.com). The point is not just to sell more seats. The reported model is closer to Palantir’s old forward-deployed playbook: put technical teams inside messy organizations, connect the model to real workflows, and make adoption somebody’s full-time job. That helps explain why Anthropic has been building the plumbing around Claude so aggressively. Its enterprise offering now emphasizes connectors into internal tools and data, deep research across company knowledge and the web, collaborative artifacts, and terminal-based coding agents that can work inside existing development environments (claude.com). In March, Anthropic also announced a $100 million Claude Partner Network for 2026, explicitly aimed at helping consultancies and service firms move customers from proof of concept to production (anthropic.com). The company is spending money not just on models, but on the people and channels that make models stick. That strategy only works if the compute is there. On April 6, Anthropic said it had signed a new agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity starting in 2027 (anthropic.com). TechCrunch reported that Broadcom filings point to 3.5 gigawatts, making it Anthropic’s biggest compute commitment yet, and said the company’s revenue run rate has climbed to $30 billion, up from $9 billion at the end of 2025, with more than 1,000 business customers spending over $1 million annually (techcrunch.com). A company does not line up that much power for chatbot subscriptions. It does it when it expects Claude to become infrastructure.