AWS and Anthropic Bet $1B on Engineers to Prove AI Pays Off

From selling models to shipping them

The pitch for enterprise AI has quietly changed. Instead of handing over an API and a demo, the big labs now send their own people in.

AWS, Anthropic, OpenAI and Google have all expanded forward-deployed engineer programs recently, according to TechRadar. The idea is simple: the model is no longer the hard part. Getting it to work inside a company's aging systems, messy data and half-defined workflows is.

The money attached to that idea is not small. OpenAI has backed its FDE business with a $4 billion investment, AWS is footing a $1 billion bill, and Anthropic says it will train 10,000 specialist engineers by the end of 2027. That last number is a training commitment, not just a hiring spree — a sign Anthropic expects demand to outrun the existing talent pool.

This is a different sales motion from the one that defined the last three years of AI hype. A license is easy to sign and easy to cancel. An engineer sitting in the customer's office for months is much harder to walk away from — for both sides.

The forward-deployed engineer hiring numbers are the real story

Lightcast's research tracks the role from a niche curiosity to a genuine hiring category in under three years.

In 2023 and 2024, US job ads for forward-deployed engineers sat at roughly 200 each year. In 2025 that jumped sixfold to about 1,200. Then, in just the first nine months of 2026, Lightcast counted 7,300 roles — another sixfold increase year over year.

That curve is steep even by AI standards, and it points at something the model benchmarks don't show: companies have bought the tools and are now paying people to figure out what to do with them.

The most interesting detail is who is hiring. Three of the four top advertisers of FDE roles are consultancies — Deloitte, Accenture and KPMG — with Google the only tech firm in that group. Anthropic's recent move to support engineer roles inside third-party consultancies fits neatly with that. The labs are increasingly happy to let the big integrators do the unglamorous work, as long as the tokens keep flowing.

Why AI vendors are paying for your deployment

There is a commercial logic here that is easy to miss. AI vendors are not being generous. They are defending revenue. If a company buys AI tooling, fails to integrate it and quietly cancels the contract, the vendor loses the account and the reference story. Sending an engineer in is cheaper than losing a logo.

The bottleneck has moved. Two years ago the question was whether a model could do the task. Now it is whether the task can be wired into a legacy system, governed properly and measured in a way a CFO accepts. That is engineering and consulting work, not research.

Wages are rising as a result. Lightcast's head of global research, Elena Magrini, told CNBC that sharp wage growth signals companies competing over a limited pool of talent. When demand for a role multiplies twelvefold in two years, the people who can do it set the price.

The catch: proof is still thin

None of this tells us whether the AI being deployed actually returns more than it costs. A billion dollars of AWS engineers is a bet on future revenue, not evidence of past ROI. The same tension runs through the whole sector — the industry has committed sums that require enormous downstream value to justify, a math problem we've broken down before.

For buyers, though, the shift is good news in the short term. Vendors competing on who will hold your hand the longest is a better market than vendors competing on benchmark scores. The risk is lock-in: once a consultancy and a lab have rebuilt your workflows around one model, switching gets expensive.

The most successful AI companies over the next two years may not be the ones with the best models. They may be the ones whose engineers are hardest to remove.

What it means for you

If you run a business in the US, UK or EU, forward-deployed engineers are already available through AWS, Anthropic, OpenAI and Google, and increasingly through Deloitte, Accenture and KPMG. Pricing is bespoke and quoted per engagement — none of the vendors publish FDE rates, and the $1 billion and $4 billion figures are internal investments, not customer prices.

The practical takeaway: ask any AI vendor what deployment support is included before you sign, and whether that support comes from the lab or a consultancy partner. The answer tells you how much of your success they are actually on the hook for.

The forward-deployed engineer is the least glamorous role in AI and possibly the most important one right now. Models are good enough to sell; they are not yet good enough to install themselves. Until they are, the companies winning enterprise deals will be the ones willing to put people on site.