The next AI fight may look less like a model launch and more like a consulting engagement.
On May 4, Anthropic announced a new AI services company with Blackstone, Hellman & Friedman, and Goldman Sachs. The company is meant to help mid-sized businesses bring Claude into core operations, with Anthropic applied AI engineers working alongside the new firm’s team.
TechCrunch reported that OpenAI is preparing a similar move through a venture called The Development Company, backed by private equity investors including TPG, Brookfield Asset Management, Advent, and Bain Capital.
So yes, the frontier labs are now trying to become services companies too.
Of course they are.
The deployment problem
Everyone talks about model capability.
The boardroom problem is deployment.
A model demo is easy. A production workflow is annoying. It has access control, internal data, approvals, audit logs, edge cases, exception handling, user training, weird legacy systems, compliance requirements, and a manager who wants to know why the AI made that decision on Tuesday.
Most companies do not have enough people who understand both the model side and the business process side.
That gap is where the money is.
This is why “AI services company” is not as boring as it sounds. It is the delivery layer that turns a model subscription into actual operational change.
Private equity wants a lever
The investor list is the tell.
Private equity firms own or influence large networks of companies. Those companies have repetitive work, messy back offices, overloaded IT teams, and constant pressure to improve margins.
If an AI services firm can walk into those portfolios with a semi-standardized playbook, the model lab gets distribution and the investors get a productivity story.
This is not just “we believe in AI”.
It is “we own companies where AI could be installed”.
That is much more concrete.
And it explains why the services layer is suddenly strategic. If OpenAI or Anthropic can help their financial backers deploy AI across hundreds of portfolio companies, the model becomes embedded before a competitor has time to run the same sales cycle.
The Palantir lesson
The forward-deployed engineer model was never just about engineering.
It was about intimacy with operations.
You put technical people close to the customer. They learn the workflow. They build around the real process instead of around a generic product brochure. The software becomes harder to replace because it starts reflecting the customer’s internal reality.
AI makes that model more valuable.
A general model needs context. A company has context everywhere, but most of it is trapped in documents, spreadsheets, inboxes, databases, meetings, and someone’s memory of why the process is broken.
The services team is the bridge.
What this means
This is another sign that frontier AI is moving past the pure product phase.
The labs still need better models. They still need cheaper inference. They still need trust.
But they also need hands.
They need people who can sit with a healthcare group, a manufacturer, a lender, or a logistics company and turn “we should use AI” into one workflow that actually saves time.
That is messy work.
It is also where enterprise software gets real.
OpenAI and Anthropic are not just racing to build the model that knows the most.
They are racing to own the path from model to business result.
That path looks a lot like services.
Sources: Anthropic, TechCrunch