Deployment is the product: what the AI labs' services ventures signal

Anthropic announced an enterprise AI services company on 4 May 2026, with Blackstone, Hellman & Friedman and Goldman Sachs as founding partners, in a venture TechCrunch reported was valued at $1.5 billion. A week later, OpenAI launched a deployment company of its own, seeded with $4 billion from OpenAI and outside investors, and agreed to buy the AI consultancy Tomoro to bring in its roughly 150 engineers and deployment specialists from day one. Two rivals, one idea, a week apart.

The idea is not new. Palantir built its business on it over two decades: send engineers into the customer's operation, learn how the work actually happens, ship working software early, and stay until the system runs in production. The industry spent years calling that model unscalable. Then Palantir's commercial results made the argument for it, and now the labs themselves are copying the structure.

Why the money moved

Two reasons, and they reinforce each other. The first is that models are becoming interchangeable at the level most businesses use them. When several providers offer comparable capability, the durable value shifts to whoever gets the capability working inside a specific company's systems, data and rules. The second is where the money sits. Enterprise software has always needed services around it, the integration, configuration and support that make it work, and that work is paid for separately. The labs were selling the software and watching the services money go elsewhere.

There is also the awkward research backdrop: MIT's widely quoted 2025 study, which found 95% of organisations getting zero return from generative AI despite $30 billion to $40 billion of enterprise investment. The bottleneck was never the model. It was everything between the model and the workflow, and that is precisely the gap a deployment business exists to close.

What buyers should take from this

The signal is straightforward. The most sophisticated players in the industry have concluded that installed, governed, production systems are the product, and that advice, licences and demos are not. A buyer can apply the same standard at any scale, whether the counterparty is a lab-backed venture or a specialist firm.

Ask what will be running in production, and when the first working version arrives; weeks is the right answer, quarters is not. Ask who does the work and where they sit relative to your systems and your data. Ask how output is reviewed before it takes effect, and what the audit trail looks like. Ask what gets measured, against what baseline. A partner with good answers is selling deployment. A partner without them is selling the demo, and 2026 is a poor year to be buying demos.

We have built our whole model on that conclusion, which is why it is pleasant, if not surprising, to watch that much money arrive at the same one.