Something shifted in May 2026. The three companies building the most capable AI models all announced, within weeks of each other, that they're getting into the services business. The kind where you send people to a client's office and help them deploy things.
Anthropic brought in the finance guys
On May 4, Anthropic spun out a separate enterprise AI services company backed by Blackstone, Goldman Sachs, and Hellman & Friedman. A research lab doesn't do that. That's a consulting firm with a model company attached.
Then the partnerships came fast. PwC alliance expanded on May 14. Stainless acquired May 18. KPMG alliance announced May 19. In two weeks, Anthropic went from "Claude is good at coding" to having relationships with two Big Four firms and a stack of deployment infrastructure.
The investor mix is the tell. Blackstone and Goldman aren't backing this because they think Claude 4 will score higher on benchmarks. They're backing it because enterprise AI deployment is a massive recurring-revenue services market that barely exists yet.
OpenAI launched a deployment arm
OpenAI is calling it "The Deployment Company." The focus is getting AI running inside critical enterprise workflows.
They also shipped realtime voice models in the API on May 7. That matters for this story because voice is how AI gets embedded in customer-facing operations. Call centers, field service, healthcare intake. The things enterprises actually want to automate but haven't been able to.
OpenAI isn't just competing on model quality anymore. They're trying to close the gap between "impressive demo" and "running in production at a Fortune 500."
Google's playing both sides
Google I/O 2026 looked different this year. Gemini 3.5 and Gemini Omni got the headlines, but the product design told a different story.
Gemini Spark builds custom apps from natural language. Daily Brief aggregates your information into a morning summary. Universal Cart lets an agent purchase things across retailers on your behalf. These are consumer features, but they're also proving grounds for the agentic workflows enterprises want.
The $100/month AI Ultra tier is Google testing whether consumers will pay enterprise prices for AI that acts on their behalf. If they will, the enterprise sales pitch gets a lot easier.
Models are becoming table stakes
Two years ago, the competition was benchmarks. Who scores highest on MMLU, who passes the bar exam, whose model writes better code.
That race isn't over, but it matters less for revenue now. GPT-4o, Claude Opus, Gemini 3.5 are all good enough for most enterprise use cases. The differentiator is whether you can get the model running inside someone's ERP system, connected to their data, with guardrails that compliance signed off on. That's the hard part.
Consulting firms as distribution
The PwC and KPMG partnerships make sense when you think about distribution. These firms already have relationships with every large enterprise. They already do transformation projects. They already have people in the building.
For Anthropic, that means instant access to thousands of enterprise clients without building a sales team from zero. For the consulting firms, it means they can sell AI transformation with a real model company behind them instead of generic proposals.
OpenAI has been doing similar deals. Google has its Cloud consulting arm. Everyone's landing on the same conclusion: whoever is already deployed when the budget shows up is the one who wins.