Amazon Bets $1 Billion on "Forward-Deployed" AI Engineers, the Same Wager OpenAI and Anthropic Already Made

A consensus is forming across the AI industry, and it has nothing to do with model capabilities. It has to do with a far less glamorous problem: most enterprises still cannot get AI to work.

On Tuesday, Amazon Web Services became the biggest name yet to put money behind that problem. AWS announced a new internal organization dedicated to forward-deployed engineers, specialists who embed directly inside customer teams to build and ship AI systems, and committed $1 billion to standing it up. The unit will deploy in pods of roughly five or six engineers running 45-day engagements, drawing from a workforce AWS describes only as numbering in the "thousands," AWS VP of frontier AI engineering and services Francessca Vasquez told CNBC and Reuters. Some of those engineers will be hired externally, with others moved over from inside Amazon.

The pitch, laid out in AWS's announcement, is that the model is "agentic-first," meaning engineers use purpose-built agents to compress deployments from months into days, and that customers are left self-sufficient once the engagement ends. The company says clients already working with the teams include the Allen Institute, Cox Automotive, the NBA, the NFL, Ricoh, and Southwest Airlines. Vasquez told Reuters there was heavy demand from customers asking for help driving agentic patterns into their workflows.

What makes the timing notable is not that AWS had the idea. It is that AWS is roughly the third major AI player to land on the same answer in the span of two months.

What a "forward-deployed engineer" actually is

The forward-deployed engineer, or FDE, is not a new role. Palantir coined the term more than a decade ago, building its business around engineers who deploy into a client (a government agency, a bank, a manufacturer) and write production code from inside the organization rather than handing over software and a manual.

The appeal is structural. An embedded engineer can navigate a client's internal politics, work with their real data and governance constraints, and respond as problems surface. Much of the underlying technology gets reused from one deployment to the next, while the implementation is tailored to each company. The client gains an injection of expertise and offloads responsibility for the rollout onto the contractor.

The catch is cost. The FDE model is labor-intensive by design. It requires maintaining a standing corps of expensive engineers to install and maintain systems, which runs against the high-margin, sell-once economics that software companies usually chase.

For most of the last decade, that trade-off kept the model mostly Palantir's. What changed is that AI deployment turned out to be precisely the kind of problem the model handles well, and the failure rate of doing it any other way became hard to ignore. Box CEO Aaron Levie captured the mood in a May LinkedIn post, predicting forward-deployed engineering would become one of the most in-demand jobs in tech. LinkedIn data cited in the same Reuters report found demand for the role grew more than fortyfold between 2023 and 2025.

OpenAI and Anthropic got there first

Amazon is following a path two of its own investments blazed earlier this year.

On May 4, Anthropic announced a joint venture to deploy enterprise AI services, with Blackstone, Hellman & Friedman, and Goldman Sachs as founding partners, plus backing from Apollo Global Management, General Atlantic, GIC, Leonard Green, and Sequoia. The Wall Street Journal, as reported by TechCrunch, pegged the venture at roughly $1.5 billion in committed capital, anchored by a $300 million commitment each from Anthropic, Blackstone, and Hellman & Friedman. The structure matters: Anthropic is a minority partner supplying engineering muscle, while the financial firms provide capital and, just as importantly, a built-in client pipeline across hundreds of portfolio companies. Anthropic CFO Krishna Rao said in the announcement that enterprise demand for Claude was outpacing any single delivery model. The venture is aimed at mid-market firms across healthcare, financial services, manufacturing, retail, and real estate that have the budget to want AI but not the in-house talent to deploy it, according to CNBC.

Within hours of Anthropic's announcement, OpenAI's own plans surfaced, and the company later formalized them as The Deployment Company, or DeployCo. It is a bigger and structurally different bet: more than $4 billion raised from 19 investors at a reported $10 billion valuation, led by TPG with Advent International, Bain Capital, and Brookfield Asset Management as co-lead founding partners, alongside Goldman Sachs, SoftBank, and consulting names including Bain & Company, Capgemini, and McKinsey, per MarkTechPost's breakdown. Unlike Anthropic's minority position, DeployCo is majority-owned and controlled by OpenAI and led by COO Brad Lightcap, AI Business reported. OpenAI also moved to buy applied-AI consultancy Tomoro, folding in roughly 150 engineers with deployment experience at companies including Tesco and Virgin Atlantic to seed the team.

Set side by side, the strategies diverge in ways worth tracking. Anthropic has paired its in-house applied-AI engineers with a partner-and-ecosystem approach, and in June it leaned further into that posture, launching a Services Track and Partner Hub in its Claude Partner Network and announcing tie-ups with systems integrators DXC and TCS to push Claude into regulated industries such as banking and airlines. OpenAI has opted for direct ownership of the delivery arm. AWS has done something different from both. Its $1 billion comes entirely off Amazon's own balance sheet, with no outside investors and no joint-venture structure, as Yahoo Finance noted.

Why everyone wants to be Palantir now

The common thread is a recognition that the money in enterprise AI may not sit in selling tokens. It sits in services.

The math is blunt. For every dollar companies spend on software, they spend roughly six on services, the ratio that built consulting into a multitrillion-dollar industry, Fortune reported. By embedding engineers rather than licensing a tool and walking away, the AI players are positioning to capture that spend directly, putting them in competition with Accenture, Deloitte, McKinsey, and the rest of the implementation establishment. As one Omdia analyst told AI Business, AI companies have looked in the mirror and decided they want to be Palantir.

There is a defensive logic too. An embedded engineer who rebuilds a client's core workflows around a particular model creates switching costs that a software subscription never could. Whoever gets inside the mid-market first accumulates case studies, proprietary deployment templates, and operational lock-in. With both OpenAI and Anthropic reportedly eyeing IPOs as soon as this fall, demonstrating durable enterprise revenue rather than impressive benchmarks has become urgent.

The skeptic's case

Not everyone is convinced the gold rush is what it appears to be. Constellation Research CEO Ray Wang has warned about "fake FDEs," engineers who function as glorified sales support rather than the deeply technical, business-fluent operators the model actually requires. The economics cut both ways. The same labor intensity that makes the model effective also threatens to drag these high-margin AI businesses toward something closer to Accenture's cost structure. Analysts have flagged exactly that risk for OpenAI's DeployCo, whose engineer headcount could climb into the thousands and carry a multibillion-dollar annual cost before a single contract is recognized.

There is a closing irony. Forward-deployed engineering is one of the few roles booming in a tech labor market otherwise defined by contraction. Amazon itself has cut more than 30,000 corporate jobs since October, even as it staffs up a unit premised on the idea that AI deployment still very much needs humans in the room.

That tension may be the most honest signal in the story. The models are good enough that three of the most sophisticated AI organizations on earth have concluded, more or less at once, that the bottleneck is no longer the technology. It is the people, the workflows, and the reality of getting it all to run. The companies betting a billion dollars each on closing that gap are betting it will not close on its own anytime soon.