This year the model labs stopped selling models and started selling delivery.
Microsoft has committed $2.5 billion and roughly 6,000 engineers, consultants and industry specialists to a unit called Frontier Company, whose job is to sit inside enterprise clients and build AI systems that produce measurable results. Amazon has put a parallel $1 billion behind the same idea. OpenAI spun out a majority-owned Deployment Company in May, raising over $4 billion from nineteen investors to do exactly one thing: get enterprises from pilot to production.
OpenAI's own chief revenue officer put the logic plainly: the constraint is no longer model capability. It is helping companies integrate AI into the infrastructure and workflows that actually run their business. That is, word for word, the job description of a systems integrator.
Why the labs want this work
Model quality has stopped being the differentiator it was two years ago. Every serious lab now has a model good enough for most enterprise use cases. What separates revenue outcomes is who actually gets the thing live, adopted, and embedded - and that has always been delivery work, not research work.
That is precisely the territory the Big Four, systems integrators and boutique consultancies used to own outright. It is also, not coincidentally, where the recurring revenue lives. A model API is a commodity with thinning margins. A five-year enterprise deployment relationship is not.
The conflict of interest nobody is naming
A delivery arm owned by the model vendor is not a neutral advisor, and it was never going to be. Its economics reward depth of adoption on its own platform - more seats, more tokens, more surface area integrated into your operations - not the best-fit architecture for your problem. That is not a criticism of the people doing the work. It is a structural feature of who is paying them and what they are measured on.
The Big Four are converging on the same structural issue from the other direction. Several are pivoting toward multi-year managed services contracts - effectively agreeing to run, not just build, a client's AI-enabled back office, with some firms projecting this could generate a fifth of consulting revenue within a few years. The party recommending the solution increasingly also owns delivering it and, now, running it indefinitely. Different vendor, same structural tension.
Three delivery models, three different trade-offs
| Model | What you get | What it costs you |
|---|---|---|
| Lab-native delivery (Frontier Company, Deployment Company) | Speed, deep platform expertise, direct access to the model roadmap | Deep lock-in to one vendor's stack and pricing; the advisor and the platform are the same company |
| Big Four / SI managed service | Scale, continuity, a single throat to choke on a multi-year contract | Your in-house capability atrophies; you're paying someone else to run what should eventually be yours |
| Independent boutique | Architecture-neutral advice, accountable only to your outcome | Smaller team, slower at pure scale, no proprietary platform access |
What we would actually ask before signing
- Does the pricing structure pay them more the deeper you go on their stack? If so, price is not the only thing you're negotiating - architecture neutrality is.
- Who owns the runbook when they leave? A lab-native or Big Four delivery team that never plans to leave has less incentive to document a clean handover than one that does.
- What happens to your in-house team's skills during the engagement? If the answer is "they watch," you are buying a permanent dependency, not a capability.
The bottom line
None of this makes lab-native delivery or Big Four managed services the wrong choice - for some programmes, speed and scale genuinely outweigh independence. But the decision has changed shape. It used to be build versus buy. It is now: who do you want owning the recommendation, the delivery, and increasingly the ongoing operation, all at once - and are you comfortable with what that concentration does to your leverage two years from now.