AWS said Tuesday it’ll spend $1 billion standing up a Forward Deployed Engineering unit that parachutes pods of five or six engineers into customer organizations for 45-day sprints, staffed eventually by “thousands” of hires. It’s the first hyperscaler to formalize the model, and Amazon is writing the check itself, off its own balance sheet, without a joint venture or an outside investor in sight.

That structural detail is the story. Anthropic set up its “AI services company” in May 2026 with Blackstone, Hellman & Friedrich, and Goldman Sachs to help midsized businesses deploy Claude models, at a reported $4 billion valuation. Days later, OpenAI stood up its own deployment vehicle with TPG, Advent International, Bain Capital, and Brookfield, valued at $1.5 billion. Google has taken a third path, committing a $750 million partner fund for agentic AI deployments rather than building the muscle in-house.

Palantir coined the phrase “forward-deployed engineer” more than a decade ago, and the term has finally metastasized into the frontier-model economy. What was once a bespoke consulting posture is now the assumed unit of enterprise AI delivery.

“The currency that the customers are always talking about right now is speed,” said Francessca Vasquez, AWS vice president of frontier AI engineering and services, in a Reuters interview. AWS points to a customer roster already running through the model: the Allen Institute, the NBA, Ricoh, the NFL, Cox Automotive, and Southwest Airlines. The NFL, Amazon says, shipped NFL Fantasy AI and NFL IQ into production in weeks.

The awkwardness is that Amazon has poured billions into Anthropic and, indirectly, sits across from OpenAI in the same enterprise deals its new unit will now chase. An AWS spokesperson told CNBC the company expects to have the opportunity to work with the OpenAI and Anthropic FDE arms rather than merely compete with them.

Read that as the tell. When the model layer starts capturing services revenue that historically belonged to the cloud, the cloud responds by insourcing the services function and framing coexistence as strategy. The forward-deployed engineer, once a Palantir affectation, is now the contested surface of the AI stack.

Sources