OpenAI files for a $10B services venture: the labs step into consulting’s lane.

In the space of ten days, the two big arguments holding up the consulting middleman case fell apart. The case rested on two ideas. First, the AI labs need partners to reach big corporate buyers, because access to the models runs through the platform relationships consulting firms already control. Second, even once buyers can reach the models directly, they still need someone to install and run them, and consulting firms own that installation work. On April 28, AWS chief Matt Garman walked onto a stage in Seattle and knocked out the first idea. On May 4, OpenAI’s Brad Lightcap filed the paperwork that took out the second.

An earlier edition documented the first blow: OpenAI models arriving on AWS Bedrock, ending eighteen months in which only Microsoft could distribute them. The distribution advantage was already gone when Bloomberg confirmed on May 11 that DeployCo, the OpenAI and TPG Capital venture announced a week earlier, is buying a consulting firm to serve as its operating core. The labs did not just make access to the models cheap and common. They are building the delivery layer that used to require a consulting middleman to reach the client.

The reason this is a pincer and not just a bad week is that a single setback leaves an escape hatch, and DeployCo closes it. When Bedrock made distribution a commodity, the firms that moved fastest still had a defensible line: the model is now table stakes, but deploying it against real corporate workflows still takes our judgment. That line survived April 28. DeployCo, a $10 billion professional services firm funded by OpenAI and TPG and built explicitly to deliver AI work, says directly that the labs do not need a middleman for delivery. The firm that built its AI practice on the premise that OpenAI needs it to reach clients now competes with a $10 billion firm that OpenAI owns. Anthropic disclosed a parallel private-equity vehicle in the same week.

What is striking is what the consulting firms are not saying. No major firm has publicly recast its AI positioning in response to DeployCo. The McKinsey and OpenAI Frontier Alliance page, announced in February 2026 as the premier path to OpenAI capability through QuantumBlack, has not changed. Accenture’s lab partnership pages are unchanged. Deloitte still leads with its Google and Microsoft relationships as the credential. The silence is rational, because a public rewrite admits the shift and invites client questions the firms cannot yet answer, but it is a delay, not a strategy.

The firms least exposed are not the ones with the deepest lab ties. They are the ones whose positioning never rested on being the middleman. Cognizant’s model of underwriting client outcomes requires an operator willing to carry client profit-and-loss risk at the account level, which the labs are not built to do. Hackett prices delivery as software rather than judgment, and the cost structure DeployCo needs makes that margin impossible at scale. Vendor-neutral governance advisory, the ISG model, is untouched because it never claimed a privileged lab relationship in the first place.

The one layer DeployCo cannot easily own is the one that already exists before it arrives. A business-logic specification, the code that encodes how a client’s process runs, how its agents decide, and how those decisions can be audited and replayed, belongs to the client the moment it ships as an artifact. DeployCo can buy delivery capacity. It cannot buy, by acquisition, the specification a client already owns. Watch three signals next: whether the Frontier Alliance is quietly recast as a research relationship rather than a sales channel, whether DeployCo starts recruiting senior partners away from McKinsey, Deloitte, and Accenture, and whether Anthropic’s private-equity structure produces its own announcement. Two labs moving into professional services in the same week is not a coincidence. It is the market deciding the middleman is a margin the labs should own.