The basketball analysts at the NBA’s league office in New York have been working with software that tracks every pass, shot and defensive rotation in real time. The system runs on Amazon Web Services, and the people who built the machine-learning models behind it will soon be sitting in the same building, ready to answer questions and adjust code on the spot. They are part of a new AWS unit that will send thousands of engineers to work inside customer companies.
Amazon announced on June 30 that it would spend 1 billion dollars to build the unit, which it is calling a forward-deployed engineering organization. The engineers will embed with customers for stretches of about 45 days, with the goal of compressing what typically takes months of integration work into a few days or weeks. The NBA and Ricoh, the Japanese office-equipment maker, are among the first customers.
The concept borrows from a playbook that Palantir popularized in government and finance, where software engineers live with clients, learn their workflows and build systems on the spot rather than delivering code from a distance. AWS is adopting the model for AI, betting that the biggest obstacle to corporate adoption is not the quality of models but the difficulty of wiring them into existing operations.
The unit will be run by Francesca Vasquez, a vice president who has led AWS’s work on frontier AI engineering and services. She said in an interview that the initial deployment will number in the thousands of engineers, with teams drawn from across AWS and trained in the specific models and tools the company now sells. The scale of the hiring has no direct precedent at the cloud giant.
The service will not be free. Amazon plans to sell the on-site engineering as part of larger enterprise agreements, with pricing tied to the scope of the engagement, and executives expect the unit to become a profit center in its own right. The company has also said it will treat the deployments as a training ground, using the projects to build reusable tools that later customers can configure without as much hand-holding.
The economics of the move are unusual for a business built on self-service. AWS grew by letting customers provision servers with a credit card, without talking to a human. Embedding engineers in customer offices is labor-intensive and expensive, and the 1 billion dollar budget signals that Amazon expects the service to pay for itself through larger cloud contracts and longer relationships.
The logic is that AI is different from earlier cloud workloads. A database or a virtual server works the same way everywhere, but an AI system has to be taught the customer’s data, processes and language. The companies that figure out how to do that quickly will capture the spending that follows, and Amazon is betting that having its own engineers on site is the fastest way to make that happen.
The market is already crowded. Palantir has built a multibillion-dollar business on forward-deployed engineers and is expanding into commercial AI. Microsoft, Google and the consulting arms of the big accounting firms all offer similar services, and the AI startups themselves, including OpenAI and Anthropic, have begun assigning engineers to their largest customers. AWS’s entry raises the stakes for all of them.
There are risks. Embedding engineers with customers is hard to scale profitably, and the model depends on the engineers actually accelerating projects rather than becoming expensive on-site support staff. AWS will also have to manage the cultural friction of teams that move between dozens of customers a year, learning a new business every six weeks.
The timing reflects a shift in how cloud companies measure success. For years, the industry competed on price per unit of computing; the next phase is competing on how quickly customers can turn that computing into working products. Embedding engineers is the most direct answer to that challenge, and analysts said other cloud providers are likely to respond with programs of their own, setting up a competition that will be fought in customer offices rather than in data centers.
Customers see the appeal. The NBA’s partnership with AWS already spans player tracking, video analysis and fan engagement, and the league has been open about wanting deeper integration rather than a catalog of separate tools. Ricoh, which is pushing into workplace services, is exploring AI systems that can answer employee questions and automate document workflows, applications that need hands-on configuration.
For Amazon, the unit is also a hedge. The company has invested tens of billions of dollars in AI infrastructure and models, and its growth now depends on customers actually using them. Sending engineers out to make that happen is expensive, but it is cheaper than building data centers that sit idle. The 1 billion dollar budget is, in effect, the cost of making sure the AI build-out turns into revenue.
