On July 2, Microsoft announced it was starting a new business. Not a product, not a feature, an entire operating company called Microsoft Frontier Company, backed by a 2.5 billion dollar investment and staffed with roughly 6,000 engineers and industry specialists whose entire job is to walk into other companies and make Microsoft’s AI actually work for them. Four days earlier, on June 6, a very different announcement had gone out under Microsoft’s name. Chief People Officer Amy Coleman told employees the company was eliminating around 4,800 roles, about 2.1 percent of its global workforce, in what she described as a refocusing of people, investment and energy toward the priorities that matter most going forward. Read those two announcements back to back and you get a company that is simultaneously telling Wall Street it is shrinking and building an entirely new business line large enough to function as a standalone company in its own right.
I do not think that tension is an accident, and I do not think it is even particularly well hidden. It is, in a strange way, the whole strategy stated plainly if you know where to look. Microsoft is not contracting. It is reallocating, aggressively, away from roles the company no longer considers central and toward a very specific bet about where enterprise AI spending is actually going to land over the next few years. Understanding why requires looking past the headline numbers and into a statistic that Microsoft, and every one of its competitors making similar moves this summer, is quietly obsessed with.
Research from MIT’s Project NANDA found that 95 percent of enterprise generative AI pilots deliver zero measurable impact on profit and loss. Companies have spent two years buying AI tools, running pilots, hiring consultants, and standing up internal task forces, and the overwhelming majority of that spending has produced nothing a chief financial officer could point to on a balance sheet. That statistic is the actual origin story of Frontier Company, far more than any press release language about transformation or outcomes. Microsoft looked at a market full of companies that bought its software and still could not make it work, and decided the fix was not a better product. It was people, thousands of them, physically embedded inside client organizations to build and operate the AI systems those clients could not get working on their own.
Read More: Everyone Is Diagnosing Themselves and No One Is Getting Better
Judson Althoff, the CEO of Microsoft’s Commercial Business, framed the new unit in a statement announcing the venture as something that goes beyond what has been labeled Forward Deployed Engineering, calling it the largest, most capable, outcome-driven engineering organization in the industry. That is a big claim, and it is worth noting that Althoff made it while resisting a label that fits his own company’s venture almost exactly. Forward Deployed Engineering, or FDE, is a model pioneered roughly two decades ago by Palantir, in which a vendor’s own technical staff sits inside a customer’s operations rather than shipping software over the wall and hoping it gets adopted correctly. Microsoft Frontier Company is that model at a scale nobody has previously attempted, but it is still that model.
The company is not alone in reaching this conclusion at nearly the same moment, which tells you something about how widely this diagnosis has spread across the industry. Amazon Web Services committed a billion dollars to its own AI deployment organization just two days before Microsoft’s announcement, with AWS vice president of Frontier AI Francessca Vasquez describing a similar approach of embedding engineers directly inside client companies. OpenAI has gone further still, founding a subsidiary called DeployCo with more than 4 billion dollars in capital, putting roughly 150 engineers directly on-site with customers. Anthropic has taken a related but structurally different path, partnering with Blackstone, Goldman Sachs and other investors on a venture aimed specifically at mid-sized companies that lack the internal resources to run AI projects themselves. Four of the most important companies in AI all arrived at essentially the same conclusion within the same several week window, that the bottleneck in enterprise AI adoption was never the models, it was the last mile of actually getting those models to do something a business could measure.
What makes Microsoft’s version notable is less the amount of money involved than the scale of the internal reorganization required to fund it, and what that reorganization reveals about how the company sees its own future. Frontier Company launched with early named partnerships including the London Stock Exchange Group, Unilever, Land O’Lakes and Accenture, giving it an immediate foothold among large, sophisticated enterprise customers rather than a slow build from nothing. Microsoft has an advantage here that its competitors mostly do not, which is that the company has already embedded engineers across much of the Fortune 500 through years of enterprise sales relationships, meaning Frontier Company is less a startup than an expansion of infrastructure that was already partially in place. Rodrigo Kede Lima, formerly president of Microsoft Asia, was named to lead the new unit, another signal that this is being run with the internal seniority of a genuine business line rather than a side project bolted onto commercial sales.
The market’s initial reaction was telling in its own right. Microsoft shares rose roughly 1.86 percent on the day of the announcement, a modest but real vote of confidence that investors saw the Frontier Company launch as additive rather than merely defensive. That reaction sits inside a much rockier backdrop, though. Microsoft stock had fallen nearly 9 percent over the thirty days surrounding the launch, and by some accounts had dropped close to a third of its value over the preceding five months, swept up in a broader wave of concern about AI’s disruptive effect on traditional enterprise software. Much of that anxiety traces directly back to Anthropic’s own rollout of increasingly capable coding and business agents earlier in the year, which rattled investors across the entire software sector by suggesting that a meaningful share of enterprise software spending could eventually route around vendors like Microsoft entirely, replaced by AI agents doing the work software licenses used to do. Frontier Company, seen through that lens, is not just an opportunistic new revenue line. It is Microsoft’s answer to a market that has started openly questioning whether the company’s core software business is as durable as it once looked.
Also Read: How Michael Truell Built Cursor, and Why SpaceX Wanted It
Satya Nadella himself addressed a related and uncomfortable question directly when the venture launched, according to reporting on the announcement, acknowledging concern that AI vendors embedding themselves this deeply inside enterprise operations could effectively absorb and commoditize the very institutional knowledge those enterprises are trying to protect. That is not a small concession. Embedding 6,000 engineers inside client companies means Microsoft’s own staff will, by the nature of the work, be exposed to proprietary processes, competitive strategy, and operational detail that a normal software vendor relationship never touches. Microsoft’s answer has been an explicit commitment that customer data and proprietary processes will not feed into shared model training, alongside assurances that clients remain free to run AI systems from competing providers even with Frontier Company embedded in their operations. Whether that commitment survives contact with the underlying business incentive, which still rewards deeper Azure dependence over time regardless of what the policy promises, is the kind of thing that will only be provable years from now, after enough client relationships have played out to see which promise actually governed behavior.
There is also a competitive positioning angle here that Althoff has been explicit about. Unlike OpenAI and Anthropic, whose deployment ventures exist to push their own proprietary models into client environments, Microsoft has framed Frontier Company as platform neutral, telling customers that data and intellectual property will not be used to train its models and that clients remain free to run AI systems from rival providers even after Frontier Company is embedded in their operations. That neutrality is somewhat theoretical in practice. Any deployment built using Microsoft’s tooling, running on Microsoft’s cloud, maintained by Microsoft’s own engineers, is going to deepen a client’s dependence on Azure over time almost by default, whatever the official policy says about openness to competitors. The neutrality claim functions less as a genuine architectural choice and more as a sales pitch aimed at enterprise customers wary of vendor lock-in, delivered by a company that stands to benefit enormously from exactly the kind of lock-in it says it is not creating.
None of this fully explains why the layoffs and the launch happened within days of each other rather than months apart, and I think the honest answer is that the timing was not a coincidence so much as a single decision expressed in two separate press releases. A company reallocating billions of dollars and thousands of headcount toward a specific strategic bet has to fund that bet from somewhere, and the cleanest way to do it is to cut the roles that do not fit the new priority at the same moment you are standing up the roles that do. Coleman’s memo to employees described the cuts as focusing people, investments and energy on the priorities that will keep Microsoft positioned to deliver for customers in a fast changing industry. Read next to the Frontier Company announcement, that sentence stops sounding like corporate boilerplate and starts sounding like a fairly literal description of what actually happened. The company shed roles it judged peripheral to the AI deployment bet and immediately redirected that capacity, and a very large amount of new capital, toward exactly that bet.
The effect on the broader enterprise software ecosystem is more complicated than a simple story of Microsoft eating everyone else’s lunch. Take SAP, the German enterprise resource planning giant whose systems run the back office of a huge share of the world’s largest companies. A Frontier Company embedded inside a client’s operations, redesigning workflows and building custom AI systems on the fly, is in some ways a direct threat to the traditional consulting relationships SAP and its implementation partners have relied on for decades. Yet rather than treating Frontier Company as a competitor to defend against, SAP appears to have concluded the opposite, expanding its RISE with SAP Acceleration program on Microsoft Azure and more than doubling the number of customers eligible for the program in 2026, while rolling out Microsoft Sentinel for SAP to provide integrated security monitoring across SAP environments. The logic seems to be that Microsoft’s embedded presence inside client organizations, whatever else it does, drives more workloads onto Azure and generates more demand for AI capabilities SAP can sell through its own tools, meaning deepening the partnership serves SAP’s interests even as Frontier Company nibbles at the edges of the consulting revenue SAP’s implementation partners used to own outright. It is a reminder that in a market this unsettled, yesterday’s competitor can become today’s dependent, and the line between disruption and partnership is often just a matter of who controls the underlying infrastructure everyone else has to build on top of.
That dynamic points to something larger happening across the entire AI services market this summer, beyond just Microsoft. For roughly three years, the public conversation about artificial intelligence centered almost entirely on the race to build the most capable model, with labs in San Francisco, London and Beijing competing on benchmarks and parameter counts while venture capital poured into companies whose entire product was a set of model weights. Frontier Company, and the nearly simultaneous moves by AWS, OpenAI and Anthropic, mark a fairly abrupt pivot away from that framing. The industry has stopped competing primarily on whose model scores highest on a reasoning benchmark and started competing on whose organization can most reliably turn that model into a documented, board-reportable improvement in a client’s actual financial results. That is a much less glamorous fight than the model race, and a much slower one, but it is also a fight with a far more direct line to revenue, which is exactly why four of the best-capitalized companies in the industry chose to enter it within weeks of each other rather than leaving the work to a fragmented layer of independent consultants and systems integrators.
Wall Street’s read on where this leaves Microsoft specifically is still unsettled. Some analysts have described the stock as trading well below fair value, with one placing Microsoft’s shares around 390 dollars against a price target north of 560, arguing that the pessimism baked into the current price does not adequately account for what a successful Frontier Company rollout could mean for long-term enterprise contract value. Others are more cautious, pointing to the capital intensity of AI infrastructure spending broadly and worrying that customer concentration among a handful of very large enterprise deployments could mute the returns on an initiative this large if even a modest share of those deployments underperform. Recent insider selling activity at Microsoft over the past several months has added a note of unease to that debate, the kind of detail that on its own means very little but that investors watching a stock down nearly a third in five months are inclined to read with more suspicion than they might otherwise.
The bigger question is whether the bet pays off, and that is where the MIT statistic becomes uncomfortable again rather than merely explanatory. If 95 percent of enterprise AI pilots are failing to move the needle on profit and loss today, Microsoft is essentially betting 2.5 billion dollars and 6,000 jobs that embedding its own people directly inside client organizations will flip that number in its favor, at scale, faster than Amazon, OpenAI or Anthropic can do the same thing with their own versions of the identical strategy. That is not a small bet, and it is not obviously going to work just because four major companies decided to make a version of it in the same month. It is, however, a much more honest admission than most of the AI industry has been willing to make so far, that the software alone was never going to be enough, and that somebody was always going to have to show up in person and make it work.
Also Read: How Michael Truell Built Cursor, and Why SpaceX Wanted It
