Vendor Strategy

Your Cloud Vendor Is Moving In

Microsoft and AWS now embed their own engineers to build and run your AI, and to own the outcome. What owners must settle in the contract first.

By Leon Soliman · 2026-07-03 · 3 min read

The vendor became the builder

On 2 July 2026 Microsoft launched what it calls Microsoft Frontier, a 2.5 billion dollar unit of about 6,000 engineers led by Rodrigo Kede Lima. Its job is not to sell you software. It embeds Microsoft people inside your operations to co-design, deploy and keep improving your AI systems, and it measures itself on your business outcomes. Named early customers include the London Stock Exchange Group, Unilever, Land O'Lakes and Novo Nordisk.

Two days earlier AWS put 1 billion dollars into its own forward deployed engineering organisation. OpenAI and Anthropic built the same motion in the spring, each pairing with private equity for capital and client access. The approach comes from Palantir, whose forward deployed engineers sit inside the client until the system works. What was once a niche consulting style is now a funded, named business at four of the largest AI vendors at once.

What you gain, and what you hand over

The appeal is real. Most companies, and almost every Mittelstand operator, do not have the engineers to turn a general model into a working part of the business. An embedded team that owns the outcome closes that gap faster than hiring ever could, and it removes the excuse that the technology is too hard to adopt.

The cost is quieter. When the pod leaves, the people who understand how your AI actually works leave with it. The system was built on the vendor's platform, tuned by the vendor's staff, and documented to the vendor's standard. You are left operating something you did not build and cannot fully see, and the natural place to turn when it breaks is the same vendor that built it.

Make the promises enforceable

The vendors have read the objection. Microsoft states plainly that customers can run models from OpenAI, Anthropic, its own labs or open source, and that customer data and IP will not be used to train models in ways that erode their advantage. That is the right posture. It is also a statement in a blog post, not a term in your agreement.

An owner's job is to turn the promise into a clause. Name who owns the code, the prompts and the operational knowledge when the engagement ends. Pin down portability, data use, retention and audit rights in writing. Keep one of your own people accountable for the system, and keep a second model wired in and tested, so that moving is a decision you can make and not a favour you have to ask for.

Frequently asked questions

Is this just a rebranded consultancy?

No. The engineers come from the vendor whose platform your AI runs on, and increasingly they take responsibility for the result. That ties the build, the day to day operation and the underlying infrastructure into a single relationship, which a traditional consultancy does not.

Microsoft says our data will not train its models. Can we rely on that?

Treat it as a commitment to verify, not a settled fact. Put the specific data, the retention period and your audit rights into the contract, so the promise survives a change of policy or personnel.

Should owners avoid these programmes?

Not necessarily. They close a genuine capability gap that many companies cannot close alone. The discipline is to keep an internal owner of the system, secure your IP and portability terms, and keep a tested alternative so you are never locked to one supplier.

The vendor that builds your AI should never be the only one who understands it.

VendorStrategy HyperscalerRisk KIStrategie AIGovernance Lieferantenmanagement Mittelstand Familienunternehmen CloudStrategie VendorLockin DigitaleSouveraenitaet Risikomanagement AIadoption ITStrategie Unternehmensfuehrung Servola