When every cloud runs every model, reading straight from the source beats building a digital twin first.

OpenAI landed on AWS on April 28, ending single-cloud exclusivity. Model access stopped being the variable. Setup time became it.

The date that changed the argument is April 28. OpenAI's GPT-5.5 and GPT-5.4, along with its Codex coding agent, went live on Amazon Bedrock, one day after the Microsoft distribution exclusivity ended. Frontier capability is now reachable from AWS, from Azure, or from the API directly, with no consulting firm required as the procurement layer. The moat that once justified making a client build a data foundation first, privileged access to the model, has simply dissolved.

In 2024 the digital twin was a legitimate precondition. The models were new, the pipelines were immature, and pointing an autonomous agent at raw material was genuinely risky. In the multi-vendor era, that same prerequisite becomes a structural liability. There is another way to ground the work: read the source document itself. The transcript, the earnings call, the contract, the board deck is the ground truth. No ontology to build, no data modernization program, no twin to synchronize before anything useful comes out.

The practical difference is cost-to-first-insight, and it is decisive. A workflow that needs weeks of plumbing before it can return an auditable recommendation does not compete with one that produces the same recommendation from an uploaded document in about a minute. Better still, the price is known before the work starts: the extraction step burns no tokens, and the single bounded model step is priced in, so the buyer sees the number up front rather than after a discovery phase. That is the conversation the CFO, now the primary buyer of this work, closes in a single meeting.

Expect a counter-move within a month or so. The firms will re-wrap the same data engagement in new language, arguing that managing many models across many clouds requires a unified data foundation. It is the same digital twin wearing a multi-vendor label. The counter-question stays the same. Can it run from the raw document today, and can the client keep running it tomorrow without you?