Anthropic stormed Wall Street with a $1.5 billion joint venture and ten ready-made financial agents, while the Pentagon signed AI deals with eight tech giants and pointedly left Anthropic off the list. The week crystallised a strange new reality: one company can simultaneously be the most wanted partner on Wall Street and the most unwanted in Washington.
The biggest move came from Anthropic, which launched two separate offensives aimed squarely at enterprise finance. First, it announced a $1.5 billion joint venture with Blackstone, Goldman Sachs, Hellman & Friedman, and a roster of other PE heavyweights including Apollo, General Atlantic, and Sequoia. The entity will embed engineers inside portfolio companies to redesign workflows and build custom AI agents — effectively an AI-native consulting firm with a direct pipeline into some of the world's largest private-equity portfolios. The day after, Anthropic released ten preconfigured agent templates for financial services: pitchbook drafting, KYC screening, month-end close, compliance review. BMO and Amalgamated Bank are the first deployments, with broader availability planned for the second half of the year. Fortune described it as Anthropic building the infrastructure to become "the operating layer for Wall Street." The announcements coincided with a Moody's data partnership and full Microsoft 365 integration — and quietly, Anthropic revealed its annualised revenue has reached $30 billion.
Washington told a different story. The Pentagon signed AI agreements with eight companies — SpaceX, OpenAI, Google, Microsoft, Nvidia, Amazon Web Services, Oracle, and Reflection — to deploy AI on classified networks. Anthropic was conspicuously absent, still frozen out after refusing to let the military use Claude for "all lawful purposes" without safety guardrails. The Defence Department has labelled the company a "supply chain risk," language previously reserved for foreign adversaries. A federal judge blocked the government's original severance order, but the chill persists. Meanwhile, Anthropic CEO Dario Amodei quietly visited the White House to brief senior staff on Mythos, the company's most powerful model, which remains restricted-access due to its cybersecurity capabilities. The contradiction — courted by the White House, blacklisted by the Pentagon — may be the defining tension of Anthropic's year.
Elsewhere in government, the Trump administration took a surprising step into pre-deployment oversight. NIST's Center for AI Standards and Innovation announced agreements with Google DeepMind, Microsoft, and xAI to evaluate frontier models before public release — a move prompted in part by the Mythos shock. It's a notable shift from the administration's prior light-touch stance and could set a precedent for how the US approaches model governance going forward. On the state level, Connecticut passed one of the nation's most comprehensive AI laws, the AI Responsibility and Transparency Act, covering employment AI, chatbot safety for minors, frontier-model whistleblower protections, and a regulatory sandbox. The governor said he'll sign it.
Cerebras filed its updated S-1 to list on Nasdaq under the ticker CBRS, targeting a $26.6 billion valuation and a $3.5 billion raise priced on May 13. The company reported $510 million in 2025 revenue, up from $290 million the year before, and its S-1 anchors heavily on a $20 billion-plus compute deal with OpenAI. It's the first pure-play AI chip company to IPO at this scale, and it's arriving at a moment when every hyperscaler is racing to diversify away from Nvidia. Separately, Bret Taylor's Sierra raised $950 million at a $15.8 billion valuation, led by Tiger Global and GV, claiming more than 40% of the Fortune 50 as customers for its AI customer agents. And ElevenLabs crossed $500 million in ARR — up from $350 million just four months ago — with a $500 million Series D backed by BlackRock, Nvidia, and Wellington.
The personal AI agent race heated up too. Meta is building an agent called Hatch — currently running on Anthropic's Claude models, ironically — designed for mainstream consumers and integrated into Instagram, where it will power agentic shopping to compete with TikTok Shop. Google is testing a personal agent codenamed Remy inside the Gemini app, positioned as a 24/7 assistant for work, school, and daily life; it also retired its browser agent Mariner, folding the technology into Gemini Agent. And Google's Gemini began rolling out to cars, with GM deploying it to roughly 4 million vehicles, plus Volvo, Polestar, and Ford following via over-the-air updates. The assistant can now answer questions using your car's actual manual — a small detail, but the kind that makes AI feel less like a demo and more like a product.