Amazon is making a $1 billion bet that the way to win the next phase of enterprise AI is to stop selling cloud and start embedding its own engineers inside customer operations. On Tuesday, AWS announced it will build a forward-deployed engineer (FDE) organization that puts Amazon staff on-site at companies like Southwest Airlines, the NFL, and the NBA to co-develop and deploy agentic AI systems, with AWS framing the timeline as "in days."
That is a quiet but consequential break from how AWS built a $100 billion-plus cloud business. For two decades, the company's pitch to enterprises was platform neutrality: rent the compute, build your own applications, hire your own systems integrator or consulting partner, and AWS would stay one layer removed from the actual implementation work. The new FDE unit pushes Amazon directly into that last-mile delivery layer, the seat Palantir carved out in the 2010s and that OpenAI and Anthropic have since copied for their own enterprise push.
The press characterization that AWS is "following Palantir's AI playbook" is editorial framing from wire coverage rather than a claim Amazon has made (Morningstar/MarketWatch). The Palantir link is useful context for what an FDE actually does, but it understates the structural shift. Palantir was a services-led company from the start; AWS is a hyperscaler that has spent twenty years explicitly refusing to be one. Walking that back, even partly, changes the relationship AWS has with the global system-integrator and consulting partner ecosystem it relies on for enterprise deals.
Amazon's framing of the unit is forward, not defensive. Francessca Vasquez, Amazon's vice president of frontier AI, said AWS will embed "thousands of experts" with customers. That headcount is a stated aspiration rather than a confirmed figure. Treat the number as a commitment language marker, not a hiring target.
The Southwest Airlines partnership, announced separately, gives a concrete shape to what the model means in practice (Amazon press release). Southwest is a long-time AWS customer with deep cloud usage; an embedded AWS team now sits between the airline and its own data, building AI capabilities the carrier historically would have built itself or handed to Accenture, Deloitte, or TCS. The NFL and NBA launches follow the same template: a customer brand, an on-site AWS team, agentic AI workloads with a "days not quarters" timeline.
There are real unknowns. AWS has not disclosed the unit economics of an FDE versus its existing Professional Services and partner-led sales engineering model, so claims that this is a faster or cheaper route to enterprise AI are analyst commentary, not company disclosure. The $1 billion figure also does not specify over what period it is spent, what portion is headcount versus partner enablement, or how much cannibalizes the AWS partner channel that already does similar work. The wire spelling of Vasquez's first name as "Francessca" appears in at least one excerpt and should be verified before any direct quote is published.
The strategic question is whether AWS can credibly run two go-to-market motions at once. The platform-neutral pitch to global system integrators says AWS is the substrate beneath everyone. The new FDE unit says AWS is also the integrator. Partners will want to know which side of that line they end up on before they route another enterprise RFP to AWS.