OpenAI launched something called the OpenAI Deployment Company this month, capitalized at more than $4 billion, and quietly acquired Tomoro — a London-based AI consultancy with about 150 engineers and a client list that includes Virgin Atlantic, Fidelity, Tesco, Mattel, and the NBA. The same week, OpenAI and Anthropic together announced $5.5 billion in joint ventures with TPG, Blackstone, Goldman Sachs, and Bain Capital. The structure is the same in both cases: instead of selling API credits to enterprises one at a time, the model providers will embed their own engineers inside the customer to actually make the thing work.
This is the most important story in enterprise AI right now, and almost nobody is covering it correctly. The press is writing it as "OpenAI gets into consulting." That framing misses the point. The model companies just looked at the services margin sitting between their API and a working production deployment — the part Accenture, Deloitte, BCG, and a thousand boutique AI shops have been billing $400/hour to capture — and decided to take it themselves.
Here's why this matters. Through 2024 and 2025, the entire AI implementation economy ran on a simple split: OpenAI and Anthropic sold the model, somebody else sold the implementation. That "somebody else" was either a Big 4 partner with a freshly minted "Center of Excellence," a Tier-2 systems integrator that pivoted from cloud migrations, or — increasingly — staffing firms placing senior AI engineers as contractors. The economics were great for the implementer because the model provider had no incentive to disintermediate them. Selling API tokens was a different business.
That changed when enterprise revenue hit 40% of OpenAI's top line and they realized the limiting factor wasn't model capability — it was customers who couldn't get past pilot. Every CIO has seen the chart. They didn't have a model problem, they had a deployment problem. The vendor that solves the deployment problem captures the relationship, the data, and the budget. So OpenAI is now buying that capability outright, and Anthropic is renting it through the world's largest PE firms.
The PE angle is the part nobody is writing about, and it's the part that matters most. TPG, Blackstone, and Bain collectively own thousands of portfolio companies. When you sign a JV with Bain, you don't sell to one CIO — you sell to a private equity ownership group that can mandate adoption across 200 companies in a quarter. The traditional Big 4 sales motion (RFP, bake-off, six-month procurement) gets bypassed entirely. The model provider becomes the default, baked into the value-creation playbook at the LP level.
If you're a hiring manager building an AI team right now, this changes your competitive set. Twelve months ago, when you posted a Senior AI Engineer role, you competed for the same talent pool as Accenture and three boutique firms. Twelve months from now, you'll be competing with OpenAI Deployment Company and an Anthropic-staffed embed paid for by your private equity sponsor. The compensation bar will rise, the loyalty model will be different, and the question "should we hire or use the vendor's people?" will get harder, not easier.
If you're a CTO who already invested in an internal AI platform team, the calculus is harder still. Your internal team's ROI was justified by the fact that hiring Accenture cost three times more. That math changes when the model provider's own engineers show up with privileged access to roadmap, fine-tuning capacity, and pricing nobody else gets.
The staffing implication is the one I've been thinking about all week. The implementation market is bifurcating fast. On one side, you have model-provider services — fewer engineers, higher leverage, deep integration with the underlying model. On the other side, you have everything else: traditional staffing, traditional consulting, and the internal teams that companies are still trying to stand up. The middle — the boutique AI consultancy charging $350/hour to "deploy enterprise AI" — is the layer that gets squeezed first, and it's already happening. Tomoro's 150 engineers didn't get acquired because they were unique. They got acquired because OpenAI needed a turnkey way to enter a market it could see closing around it.
For the engineers themselves, the play is changing. Two years ago, the highest-leverage move was joining the model lab. A year ago, it was joining the boutique that deployed for the model lab. Today it's joining the model lab's deployment arm — because that's where the model access, the customer data, and the equity upside all converge. The talent will flow there. It already is.
What does a hiring manager actually do with this? Three things. First, stop treating "AI implementation talent" as one market — it's now two, and the model-provider side will price you out for the top decile. Second, if you're hiring an AI consultancy or staffing firm to deploy something, ask hard questions about whether they survive the next eighteen months — many of them won't. Third, watch your private equity sponsor's preferred-vendor list, because if Bain or TPG signed a deal with Anthropic, your "AI build vs. buy" decision was just made above your head.
The model companies just ate the consultants. The question worth sitting with is whether you want to be the customer, the competitor, or the next acquisition.
What are you seeing? Has your PE sponsor mandated a model provider yet? Reply and tell me — I'm tracking this across our client base and the picture is changing fast.