OpenAI is putting $4 billion behind forward-deployed engineers, AWS is committing $1 billion, and Anthropic plans to train 10,000 specialists by the end of 2027. The reason is a shift in enterprise AI: model capability is no longer the main limit, while legacy infrastructure and weak use cases block returns. Vendors now take responsibility for deployment, not just licenses, which matters because adoption success decides future contracts.
Why vendors pay for on-site implementation
Forward-deployed engineer programs are expanding at AWS, Anthropic, OpenAI and Google, with vendors placing specialists inside customer teams to connect products with operational problems. Lightcast counted around 200 FDE job ads in the US in 2023 and a similar number in 2024, then about 1,200 in 2025. In the first nine months of 2026 the count reached 7,300, another sixfold rise year over year. The sequence shows demand moving from experimentation with models to applied installation work.
The mechanics are straightforward: an engineer works on the client side, maps processes, adapts the model to existing systems and data, and brings a pilot to measurable use. That approach addresses two barriers named in the report, outdated infrastructure and unclear real-world applications. Instead of handing over access and documentation, the developer shares accountability for the outcome. For buyers, the purchase increasingly includes labor for integration, not only software capacity.
The context is a change in how AI is sold, from product delivery to proof of value inside operations. Consultancies have moved quickly into the gap: Deloitte, Accenture and KPMG rank among the four top advertisers of FDE roles, alongside Google as the only tech firm in that group. Anthropic has moved to support engineer roles inside third-party consultancies. Lightcast Head of Global Research Elena Magrini told CNBC that sharp wage growth points to competition for a limited talent pool.
What deployment-led sales mean for buyers
For companies adopting AI, vendor-supplied implementation can shorten the path from pilot to production, especially where internal teams lack integration experience or data preparation capacity. A small firm may get access to skills it could not hire, while a large firm with legacy systems may use external engineers to unblock specific processes. The practical difference appears in project setup: timelines, access to systems, and success metrics become part of the commercial discussion from the start.
The same model creates selection risks that deserve attention before signing. Availability of engineers, their mandate, and handover terms affect whether knowledge stays inside the company after the team leaves. Wage pressure and scarce talent mean buyers should clarify staffing, duration, and responsibility for maintenance. This news alone does not prove returns across industries; it shows vendors are willing to fund the work required to demonstrate them.
The marker to watch is whether FDE hiring keeps growing into 2027 and Anthropic reaches its 10,000-engineer training target. If consultancies and labs sustain that staffing, deployment capacity becomes a competitive factor alongside model quality. Business buyers will then compare vendors less on benchmarks and more on delivered operational results.
