All insights

Strategy

Your AI Supplier Is Also a Financing Exposure

A contract describes the service you buy. It does not describe how the supplier pays for the infrastructure behind it, or what changes for you if that financing tightens.

Ronny BoesingPublished 6 min read

Share this perspective

Original conceptual illustration: a bridge across a misty valley at sunrise rests on several pillars that narrow into one shared foundation beneath the water. Not a real supplier, system or data.
Original vector editorial illustration · conceptual, not documentary evidence

A finance team is asked to approve a three-year commitment to an AI platform that will sit inside order handling, customer service and reporting. The demo works. The price per seat is acceptable. The security questionnaire comes back clean. The question nobody on the approval list has been asked is a different one: if the supplier's cost of funding rises sharply, or a provider further down its supply chain changes terms, what happens to this workflow?

This is a hypothetical approval scenario, not a client case and not a claim about any named supplier. It frames a bounded buyer decision: commit as proposed, commit with dependency safeguards, or keep the workflow on a smaller, reversible footing until the dependency is understood.

What the ECB actually said on 18 November 2026

The European Central Bank's November 2026 Financial Stability Review is a macro-financial assessment for the euro area. It is not procurement guidance, and it does not rate individual companies. It is still useful to buyers because it describes the conditions under which AI services are currently being financed.

Three of its observations matter here. First, the Review states that investment in the AI buildout "is highly concentrated among a small group of US firms, including hyperscalers and manufacturers of semiconductors," and that free cash flows "turned negative for most US hyperscalers in 2026, as capital expenditure on AI infrastructure exceeded operating cash flows," leading these firms to turn increasingly to external equity and debt (Box 3). Second, it notes that "circular financing is making intrinsic demand hard to gauge." Third, it links operational risk to "heavy reliance on a small cluster of third-party providers" and, in its cyber special feature, says euro area banks "rely heavily on concentrated third-party providers."

Those are the ECB's statements about the financial system. They are not a forecast that any provider will fail, raise prices or withdraw a service. The Review itself presents a two-sided view: AI investment is supporting growth, and the technology "may turn out to be just as transformative as expected." A buyer should read it as a reason to ask better questions, not as a signal to avoid AI.

Why financing belongs in a supplier decision

Most supplier assessments look at the supplier you sign with. The AI services a business buys usually sit on a chain: an application vendor, a model provider, a cloud or data-centre operator, and chip and energy supply underneath. The ECB describes the upstream layers as capital-intensive, concentrated and increasingly debt-funded, with higher yields raising debt service costs "particularly for more leveraged segments of the AI ecosystem that rely on floating-rate or private credit financing."

For a buyer, that translates into three ordinary commercial risks rather than an exotic financial one:

  1. Price risk. Usage-based prices can change when the provider's own costs change. A workflow that is economic at today's token or compute price may not be at a materially higher one.
  2. Continuity risk. A product line, model version or regional capacity can be withdrawn, consolidated or re-tiered if the provider's priorities change, including after a merger, refinancing or restructuring of a smaller vendor in the chain.
  3. Concentration risk. If several of your critical suppliers depend on the same small set of upstream providers, diversifying at the application layer may not diversify the actual dependency.

None of these requires a market crisis. They are the normal consequences of depending on a supplier whose economics you cannot see and do not control.

What regulators already expect from financial firms

Financial institutions in the EU are already required to manage this kind of dependency. The Digital Operational Resilience Act (Regulation (EU) 2022/2554) has applied since 17 January 2025 (Article 64). Article 28 requires covered financial entities to manage ICT third-party risk and, for ICT services supporting critical or important functions, to put in place exit strategies (Article 28(8)). Article 29 requires a preliminary assessment of ICT concentration risk before contracting for such services. From Article 31 onwards, the Regulation sets up an oversight framework, under which the European Supervisory Authorities designate critical ICT third-party service providers. The Financial Stability Board's December 2023 toolkit on third-party risk management is explicitly non-binding. It offers financial institutions and authorities tools for identifying critical services, mapping third-party service relationships, managing risks in service providers' supply chains, business continuity and, where possible, documenting and testing exit strategies.

Most businesses reading this are not covered by DORA. The point is not that they must comply with it. It is that regulators who have studied the problem settled on a recognisable set of practices — know which services are critical, know what they depend on, and have a workable plan if they stop — and those practices are proportionate for any organisation putting a core workflow on an external AI service.

The counterarguments are real

There are good reasons not to over-weight this risk.

The largest providers are among the best-capitalised companies in the world, and a buyer is unlikely to out-analyse the bond market on their solvency. Competition is also working in buyers' favour in places: the ECB itself notes that cheaper, open-weight models may gain traction, which could lower prices rather than raise them. Switching between model providers is, for some workloads, a configuration change rather than a migration. And excessive caution has a cost of its own: a business that refuses to commit may forgo real productivity gains while competitors take them.

These points narrow the question rather than remove it. The relevant risk is rarely that a giant provider disappears. It is that a specific price, model, region or intermediate vendor your workflow depends on changes faster than you can adapt — and that you discover the dependency only when it does.

Buyer decision checks before committing a core workflow

Before approval, the accountable owner should be able to answer these in writing:

  1. Criticality. Which business process stops or degrades if this service is unavailable for a day, a week or permanently? Who decides that, and is it documented?
  2. Dependency map. Which upstream model, cloud and region does the service rely on? Does the supplier disclose it, and will it notify you of changes?
  3. Concentration. Do your other critical suppliers rely on the same upstream providers? If so, are you actually diversified?
  4. Price exposure. What happens to the business case if usage-based prices rise materially? Is there a contractual cap, notice period or fixed-price option?
  5. Continuity terms. What notice does the contract give for model deprecation, service withdrawal or change of control? Is there a right to terminate without penalty in those events?
  6. Data and configuration portability. Can you export your data, prompts, configurations and logs in a usable format, on a timetable you could actually execute?
  7. Exit path. Is there a tested or at least credible alternative — another provider, an open-weight model, or a manual fallback — and an estimate of how long the switch would take?
  8. Review trigger. What signal would cause you to re-review the dependency: a price change, a deprecation notice, a supplier refinancing or acquisition?

If those answers exist and are acceptable, committing as proposed may be entirely reasonable. If several are unknown, the narrower option — a shorter term, a reversible scope, or a documented fallback — is usually cheaper than discovering the dependency later.

What this article does not claim

It does not assess the financial health of any named company, predict a market correction, or suggest that AI investment is unsound. The ECB passages quoted are its system-level assessments as of November 2026. The regulatory references describe obligations for in-scope financial entities and are used here only as examples of established practice. The approval scenario is hypothetical.

A practical next step

If you need to frame which workflow is critical and what a sensible first commitment looks like, start the free Workflow Fit Diagnostic. It can help clarify the business problem and a possible next step; it does not assess supplier finances, review contracts or certify resilience. The free result may be enough.

Sources & perspective

The recommendations in this article are Boesing’s analysis. These publications provide context, not independent validation of business results. Reading an original source does not verify its claims.

Share this perspective

Reader discussion

Share a considered perspective. Every comment is reviewed by a moderator before it appears; nothing is published automatically.

Use a name or initials. Please do not enter an email address.

Plain text only. Please do not share confidential or client information.

0/2000 characters

We store your display name, comment and a salted abuse-protection key that is cleared after 30 days — never your raw IP address. Guest emails are not requested.

Comment submission is not open yet: the moderation service is awaiting launch. Nothing you type here is sent or saved.

A data centre corridor with open doors revealing a sunrise over mountains and water, suggesting a clear way out.
Published insight

Strategy

A Right to Switch Is Not an Exit Plan

Regulators are making it easier to leave a cloud supplier. They cannot decide whether your organisation is ready to leave. That question belongs in the buying decision.

Ronny BoesingPublished 7 min read

Editorial illustration: a blue iceberg reveals a much larger submerged structure beneath the waterline, representing hidden operating costs.
Published insight

Economics

The Hidden Cost of AI Automation

The licence fee is the visible part. The real cost of automation sits in integration, exceptions, oversight and the work of keeping it right.

Ronny BoesingPublicly available since at least 4 min read