Amazon Web Services stood up a $1 billion Forward Deployed AI Engineers organization on June 30, embedding engineers inside customer companies to build and hand off purpose-built agentic systems that run in each client's own AWS environment. Frontier AI VP Francessca Vasquez, who leads the effort, framed it as a way to leave customers with lasting AI skills instead of just delivered software.
Palantir's playbook, now with hyperscaler scale
The Forward Deployed Engineer model was pioneered by Palantir: send engineers to sit inside a client, hunt for high-value use cases, and stay long enough to hand-carry the deployment into production before rotating out. AWS is doing the same thing at hyperscaler scale, promising fast engagements, purpose-built agents and internal workflows the customer can extend without the contractor in the room. The $1 billion is Amazon money, not outside capital, and the org is being built as an internal operating unit rather than a joint venture.
The four-way FDE race
AWS is the fourth major AI lab or hyperscaler to formalize an FDE unit in eight weeks. OpenAI and Anthropic launched $4 billion and $1.5 billion FDE joint ventures in May with backing from Goldman Sachs, Blackstone and Hellman & Friedman. Two days after AWS's announcement, Microsoft launched Frontier Company, a $2.5 billion, 6,000-person deployment unit that resists the FDE label but sells the same service.
Why the money is on engineers, not models
Roughly 95% of enterprise AI pilots deliver zero measurable business result — a statistic every vendor now cites when pitching FDE. Vasquez's team is pitched at the same gap: fast time-to-first-agent inside SageMaker and Bedrock, then a clean handoff so the customer can compose new agents against their own data-center CPU and GPU footprint. If AWS's FDEs can turn Amazon's already-huge enterprise install base into repeat agentic deployments, the $1 billion outlay pays for itself in cloud consumption alone.
Reporting based on coverage from TechCrunch, AWS's own announcement and TheNextWeb.