Microsoft spends $2.5 billion to establish "Frontier Company" and will send 6,000 engineers to client offices to make AI truly come to fruition.

Microsoft is pouring $2.5 billion into Frontier Company and appointing 6,000 engineers, who will be directly dispatched to customer offices. The goal is to turn a pilot program into measurable results, directly competing with similar setups by Amazon, OpenAI, and Anthropic.
(Recap: Even Meta has started to ask 6,000 employees to use AI sparingly—usage doesn’t equal effective output.)
(Background: Taking on Nvidia’s dominance! Rumor has it that Anthropic is teaming up with Samsung to develop customized AI chips, and Claude’s underlying compute will undergo a major overhaul.)

On July 2, Microsoft announced the establishment of a new business entity, “Microsoft Frontier Company.” It will invest $2.5 billion and staff 6,000 industry and engineering specialists, with Rodrigo Kede Lima—former Microsoft Asia president with 30 years of industry experience—serving as president.

The objective is to place the engineers themselves right beside the customers’ desks. The bet is based on the premise that what enterprises truly lack has never been smarter AI; rather, it’s people willing to spend time so that AI can genuinely be deployed and produce results.

Move the monetization battleground into customer offices

The core of this model is called forward-deployed engineering (FDE for short).

Put simply, Microsoft is no longer just selling the model and the API and calling it a day. Instead, it directly sends engineers to embed inside the customer organization—working with the customer team to design, deploy, and continuously optimize bespoke AI applications—so that testing projects are truly advanced into measurable business outcomes.

Althoff’s exact words are that this organization “goes beyond what the industry currently calls forward-deployed engineering,” and will be “the engineering organization with the largest scale in the industry, the strongest capabilities, and a results-oriented focus.”

Two platforms support this approach: an intelligence platform that allows enterprises to lay the foundation with their own proprietary data, professional expertise, and decision-making processes, and to freely choose OpenAI, Anthropic, Microsoft in-house models, or open-source models—so they are not locked into a single vendor; and a trust platform that handles observing, governance, management, and protection of the entire AI solution.

There is also a red line that Microsoft claims is non-negotiable in the middle of it all: customer data, intellectual property, and competitive advantages will not be used to train models, and therefore will not turn into Microsoft’s own products. CEO Satya Nadella’s words were quoted directly in the announcement: “There is no social license that would allow an AI future to consume the company’s intelligence that it is deployed into.”

The first customers include the London Stock Exchange Group (LSEG), Unilever, Land O’Lakes, and Novo Nordisk; and it has brought in partners including Accenture, Capgemini, EY, KPMG, and PwC.

Why rely on people, not models?

The answer is hidden in an industry fact that everyone understands: more than 90% of enterprise AI projects die at the stage of “build it but it doesn’t get used.”

No matter how strong the model is or how impressive the benchmark scores look—if no one inside the enterprise connects it to existing systems, processes, and data, then it’s only a very expensive demo. Microsoft’s logic is that AI deployment ROI (how much measurable output the enterprise gets back after investing in AI—for example, saved labor hours or improved deal conversion rates) is never something produced by a model alone; it is the result achieved only when engineering time, industry knowledge, and the customer’s internal politics are combined.

This also echoes the assessment that the longtime commentator on the subject has long emphasized: Benchmark is the starting line, while deployment is the finish line. In the past two years, the AI industry competed on model scores, but what truly determines who can get budget from enterprises is who can turn papers into numbers visible on financial statements.

What Microsoft is fighting this time is not a model war, but a “who can truly deliver AI” war. The weapon is not GPUs—it’s the number of engineers’ headcount.