Meta Wants Your AI Business. Keep Your Options Real.

A new supplier does not automatically create a usable alternative

The OECD's July 10 policy brief describes a market with competing forces: dynamism in foundation models alongside structural risks across the wider AI supply chain. Its discussion of vertically integrated firms explains how bundling and control across layers can increase switching costs. That is market analysis, not a finding about Meta's new platform. [3]

My interpretation for buyers is straightforward. A longer supplier list matters only when the organization can act on it. If changing the model also means rebuilding the workflow, reconnecting business systems and retraining the people who run it, the next sales presentation is an option with an implementation project attached.

That does not make integration a bad purchase. It makes the boundary of the purchase important. Decide which parts of the workflow benefit from a common platform and which business assets should remain usable outside it.

Follow the work out of the demonstration

Consider a hypothetical hotel company preparing to open several properties. It wants an AI-assisted workflow that assembles opening-readiness summaries from project updates and approved operating requirements. A manager reviews the exceptions before an opening decision. This is an illustrative use case, not a reported Meta deployment.

The first demonstration looks promising. The summary is clear, the exceptions are easy to find, and the operations director can picture using it every morning. Then she asks the team to take one property's workflow to a second environment.

In our scenario, the documents export easily. The harder questions concern the rules that say which exception matters, the connections to project records, and the examples used to judge whether a summary is good enough. A folder of documents has left the building. Has the working capability left with it?

This distinction is why I would ask for a small transfer exercise during evaluation. The objective is to discover what the business can carry forward and what it would need to recreate. A downloadable archive is evidence of an export. A usable replacement workflow requires a different demonstration.

The friction is bigger than the file format

There is relevant evidence outside the AI launch cycle. In its September 16 announcement summarizing a cloud-market study, Spain's competition regulator, CNMC, identifies interoperability and portability barriers, as well as commercial practices such as discounts and data-extraction charges, that can reinforce dependence. These are findings about cloud services in Spain, not measurements of enterprise AI switching or Meta customers. [4]

An OECD infrastructure paper published November 14, 2025 adds another useful distinction. Section 3.5 discusses proprietary interfaces and contractual barriers, while recognizing the convenience and efficiency integrated ecosystems can offer. It also cautions that switching barriers need not result from illegal conduct. [5]

Applied to the hotel example, these sources suggest examining technical work and commercial terms together. Can the company retrieve the relevant records? What rights does it have to reuse its configurations? What continuing commitments would remain after the workflow moved? Those are questions for the proposed agreement and implementation, not assumptions about any supplier's terms.

The sensible comparison is between the value of the integrated service and the effort required to retain a workable alternative. Requiring everything to run everywhere would itself become a project. The board needs a business case for the flexibility it is buying.

Buy enough flexibility to make a decision

I recommend three steps for the next platform evaluation. They apply the OECD and CNMC analysis to a buying process; they are proposed management practices, not a tested migration method or a guarantee of savings. [3][4][5]

• Choose a consequential slice of work. For the hotel company, use one property's readiness review, including an awkward exception. Agree what a useful result looks like before comparing environments. This keeps the portability question attached to a business outcome.
• Price the transfer. Have the team identify reusable documents, business rules and evaluation examples, then record connections and configurations that need rebuilding. Ask procurement to confirm export rights, remaining commitments and any relevant charges. Mark unknowns explicitly.
• Make the dependency decision visible. Compare the integrated option with a feasible alternative, including staff effort and service disruption. Accept dependencies that earn their place. Revisit the choice when a contract renews, a material workflow changes or another offering addresses an unmet need.

The useful answer may be to stay

Back at the hypothetical hotel company, the exercise need not end with a vendor change. The team might conclude that rebuilding is expensive and the current service is worth keeping. Or it might find that maintaining its own operating rules and evaluation examples preserves enough flexibility without duplicating the whole system.

Either answer is more useful than assuming that another logo on a shortlist creates bargaining power. The executive can now explain what the company is buying, what it depends on and what changing course would involve.

Meta's announcement is an invitation to reconsider the available suppliers. Use it to ask for evidence about the work itself. An option becomes strategically useful when the organization knows what it would take to exercise it.

Questions for executives

1. Which business rules and evaluation examples would we retain if we changed this platform?
2. What would have to be rebuilt, and who has estimated the effort?
3. Which dependencies are justified by value, and what would cause us to reconsider them?