The firms that bet their AI practices on Microsoft are now defending a moat that no longer exists.

Azure fluency was the entry ticket to enterprise AI work. When any buyer can reach the same models directly, the ticket stops being scarce.

For most of the past eighteen months, building an AI practice around Microsoft Azure was a sound bet. If OpenAI ran only on Azure, then Azure fluency was the entry point to nearly every serious enterprise AI engagement. Accenture, EY, Deloitte, and McKinsey all deepened Microsoft-aligned AI teams on exactly that logic. It was not a mistake. It was the correct read of a market where the model sat behind the cloud and the cloud sat behind the firm.

That structure is what the April 28 announcement dismantled. When a buyer can reach OpenAI's models directly on Amazon Bedrock, on Azure, or through an ordinary interface, the consulting firm is no longer the layer that grants access. In plain terms it was a distribution arrangement: part of the firm's value rested on standing between the client and a model the client could not otherwise reach easily. Remove the scarcity and that portion of the value goes with it.

The firms that hedged early look smartest now. At Google Cloud Next in late April, Google committed a 750 million dollar fund to its partners and expanded agentic AI alliances with Deloitte, BCG, and PwC around its Gemini Enterprise line. Deloitte stood up a dedicated practice and said it would grow its internal Gemini seats fourfold. Those are real repositioning moves. But notice what the reframe becomes: the firm sells itself as the neutral guide who helps the client navigate a newly complicated, multi-vendor world. The wrapper is new. The economics, billing the client for the firm's continued presence rather than for something the client keeps, are unchanged.

The tell to watch is the next round of briefings. Expect the pitch to shift toward multi-vendor strategy and toward the question of how you govern all these agents. Governance is a legitimate concern. Sold this way, though, it is the same play at a new layer: a reason the client should keep paying for interpretation rather than own the procedure that encodes the decision and can be reviewed on its own.