Microsoft Just Spent $2.5 Billion to Admit Something: Most Companies Are Bad at Actually Using AI.

Here's an uncomfortable truth the AI industry doesn't say out loud often enough: buying a model subscription doesn't automatically produce a return. Most enterprises have moved past the "let's experiment with AI" phase and landed somewhere much less comfortable, stuck trying to prove the investment was worth it.

Microsoft's answer is a new business unit called Microsoft Frontier Company, backed by $2.5 billion and staffed with 6,000 engineers and industry specialists who will embed directly inside client organizations. Not consultants who show up for a workshop and leave a slide deck. Engineers who sit inside the business and build the thing.

CEO of Microsoft Commercial Business Judson Althoff is positioning this as bigger than the "Forward Deployed Engineer" model that companies like Palantir made famous, calling it the largest, most capable outcome-driven engineering organization in the industry. That's a big claim, and it comes with a genuinely interesting tension attached.

Microsoft already has deep Forward Deployed Engineering partnerships with Accenture, Capgemini, EY, KPMG, and PwC, the same consulting giants that have built entire practices around exactly this kind of AI implementation work. So is Frontier Company a complement to those partnerships, or is Microsoft quietly building the capability to route around them? The blog post doesn't really answer that, and it's the question every one of those partners is probably asking right now.

The philosophy driving this is what Althoff calls Intelligence plus Trust: a company's proprietary data, workflows, and decision-making processes need to compound over time on a platform that also lets them govern, secure, and audit how AI touches every layer of their stack. In practice, that means a continuous improvement loop, not a one-time deployment.

The early proof points sound promising. LSEG is using this approach to let finance professionals query structured and unstructured data through LSEG Workspace. Land O'Lakes, Unilever, and Novo Nordisk are cited as other active transformations. Notably absent: hard numbers. No percentage improvement, no dollar figure, no timeframe. Just directional language like "measurable outcomes" without the actual measurements attached.

There's also a pointed message buried in here for the AI industry as a whole. Satya Nadella reportedly said there's no societal permission for an AI future that consumes the intelligence of the very companies it's deployed inside. Translation: if your AI vendor is quietly using your proprietary data and workflows to make their own models smarter, that's not a partnership, that's extraction. Microsoft is positioning Frontier Company as model-agnostic, letting customers run OpenAI, Anthropic, Microsoft's own models, or industry-specific ones without lock-in.

Whether that promise holds once 6,000 Microsoft engineers are embedded inside your operations is the kind of thing you'll only really know a few years in.

Is this genuine customer protection, or the biggest company in enterprise software finding a new way to get closer to your data?