Microsoft's $2.5 billion Frontier Company puts its own engineers inside customers to run their AI

Microsoft on Thursday announced the Microsoft Frontier Company, a $2.5 billion effort to put its own engineers inside customer organizations and build their AI systems for them. It is not a separate legal entity. A spokesperson called it "a purpose-built company with its own leadership and financial accountability" but would not call it a standalone company, and said the group of "more than 6,000 industry, engineering and AI professionals" is drawn mostly from teams that already work at Microsoft. It will be run by Rodrigo Kede Lima, until recently the president of Microsoft Asia.

[]

Strip the branding and this is forward-deployed engineering, a practice with a boring definition: instead of selling you software and leaving, the vendor sends its own engineers to sit inside your company, figure out where the technology fits, and wire it into how you actually work. Palantir built its business on the model two decades ago. Over the past two months it has become the thing every large AI provider is racing to copy.

The same bet, placed four times since May

Microsoft is the fourth of the big AI providers to stand up a deployment arm, and the last to arrive.

Judson Althoff, who runs Microsoft's commercial business, wrote that Frontier "goes beyond what has been labeled as Forward Deployed Engineering" and would be "the largest, most capable, outcome-driven engineering organization in the industry." That is the claim. The measurable parts of it are the dollar figure and the headcount, both larger than the other three. The "goes beyond" is marketing until there is a customer outcome to point at.

Why every one of them wants this now

The reason is not complicated, and the companies are fairly open about it. Businesses bought ChatGPT, Claude, Gemini, and Copilot, ran some impressive demos, and then discovered that a demo does not reorganize a company that has its own data, its own rules, and its own habits. The gap between "the model can do this" and "our operation now does this" is where the AI payoff has been getting lost.

"Having the model alone doesn't change your workflows or how you operate," Marc Nachmann, Goldman Sachs' global head of asset and wealth management, told CNBC about the Anthropic venture. "You need people who can combine the technology with what's actually happening in the business and implement those changes."

There is also a business logic underneath the customer-service language. Models are getting cheaper and more alike every month, which makes the model itself a weak place to earn a margin. Selling the labor of making the model pay off is a much larger market, and it locks in demand for all the cloud capacity these companies are spending hundreds of billions of dollars to build. Selling services, not software, is where the money moves next.

What is actually new here, and what is not

For Microsoft specifically, the honest answer is: less than the launch suggests. The company already runs Industry Solutions Delivery, the in-house arm that absorbed the old Microsoft Consulting Services, with thousands of people who build and deploy technology inside customers. It already has FastTrack for rollouts, a dedicated forward-deployed practice with Accenture, and a $1 billion, five-year alliance with EY announced in May. Frontier gathers that existing muscle under one name and a bigger budget. Some consulting roles, meanwhile, are expected to be caught in a round of layoffs Microsoft is planning for next week.

The more interesting tension is in Microsoft's own pitch. It is selling Frontier partly on model freedom: run OpenAI, Anthropic, Microsoft's own, or open-source models, and swap one for another without losing the institutional knowledge you have built. Satya Nadella has made this his test for whether a company still controls its future. "The last thing any of us want is a world where every company across every sector is ceding value to a few models that eat everything they see," he wrote in a June essay.

The catch is that model portability is not the same as vendor portability. Once Microsoft's engineers have built your systems, those systems tend to run on Microsoft's cloud and Microsoft's surrounding tooling. You may be free to change the model in the box. Getting out of the box is the part that stays hard, and it is the part the $2.5 billion is quietly designed to make harder.