In one stretch of days, AWS put $1 billion into an organization that installs AI inside other companies, and Microsoft stood up a $2.5 billion one staffed with 6,000 engineers. AWS announced its Forward Deployed Engineering org on June 30. Microsoft’s Frontier Company landed two days later. I already wrote the seller’s read on the Microsoft move, that the deployment work, not the model, is where the durable money sits. That still holds. But watching all five big AI names pile into the same idea in a single quarter, the document worth rereading is the one from May that nobody connected to this.
On May 4, OpenAI and Anthropic each built their deployment arm as a private-equity joint venture. OpenAI’s is called The Deployment Company, valued at $10 billion, funded with more than $4 billion from 19 investors including TPG, Brookfield, Advent and Bain. Anthropic paired with Blackstone, Hellman & Friedman and Goldman Sachs for a $1.5 billion version. And OpenAI guaranteed its private-equity backers a 17.5% annual return over five years.
You don’t guarantee a return on a services business
Read that term again, because it carries the whole story. A 17.5% floor over five years is what a fixed-income product looks like, not an operating partnership, and several PE analysts told Bloomberg exactly that. Consultancies don’t promise their investors a number. Services revenue is lumpy, headcount-bound and margin-thin; you can’t underwrite a guaranteed return on it. So OpenAI wasn’t buying a consulting arm when it signed that term. It was paying, up front and expensively, for something a consulting arm can’t deliver.
The asset is the org chart above the engineers
Here’s what the money bought. TPG, Blackstone, Bain and the rest own thousands of portfolio companies between them, and a PE owner can do the one thing every enterprise software seller dreams about: tell the CEO to adopt, from the board seat, and mean it. The engineers are the visible part. The distribution is the part you pay a 17.5% guarantee to lock up, a standing channel into hundreds of companies where the adoption decision comes down from the top instead of clawing its way up from a pilot.
I’ve sold on both sides of that line. At Oracle and Fastly, the deals that died slow were the ones where a champion three levels down loved us and nobody above them had signed up for the outcome. The deals that moved had air cover from the top. The hardest thing to manufacture in enterprise sales is a mandate. OpenAI and Anthropic just rented thousands of them.
Two things can be true
The deployment work is real and valuable. That isn’t in question, and it’s the same reason the value in AI keeps sliding from the model to the layer that makes it work. Most companies can’t get AI into production on their own, and someone will get paid well to help. But the reason these five moved in a quarter, and the reason two of them wrapped it in PE money, is that the deployment shop is also the delivery vehicle for the thing that decides an AI company’s revenue: whether the customer keeps using it. The services get them in the door; the deeper the team embeds, the harder the model underneath is to replace.
What it means when your competitor arrives with a board mandate
If you sell AI into the enterprise, price this in now. Your next competitive deal may not come from another vendor’s AE. It may arrive as a directive from a portfolio company’s owner, with engineers already on-site and a model already chosen. You can’t out-demo a mandate. What you can do is get to the same altitude: sell the economic buyer on a specific outcome the generalist pod won’t go deep enough to own, and make yourself the person accountable for that number. If you’re on the buying side and the mandate is landing on you, take the help, but put portability and ownership in writing, whose model, whose data, whose runbook the day the embedded team rotates off. A guaranteed 17.5% has to come from somewhere, and it isn’t coming out of the vendor’s margin.