An AI product can look like a single service while depending on several businesses. The company issuing your invoice may use another provider's model, someone else's computing capacity and a separate integration partner.
For a founder, that matters most when the service becomes part of a promise to a customer. Who can keep that promise if one of the underlying arrangements changes?
On 16 November, Federal Reserve Governor Michael Barr described risks associated with the financing behind the AI buildout, including supplier investments in smaller AI companies that then buy computing services from those suppliers. He noted that smaller recipients would ultimately need paying-customer revenue to replace that funding. These were financial-stability observations, expressed as his own views, rather than a finding that a particular supplier would fail. [1]
Reuters separately reported his warnings about AI's implications for markets, cybersecurity and customer protection. The commercial lesson for an SME is narrower: useful technology and a dependable supplier relationship require different checks. [2]
Draw the dependency map
Start with one important workflow: preparing a customer quotation, processing a document or supporting a service team. Identify the contracted supplier, underlying model and hosting arrangements where disclosed, and the integrations needed to produce a usable result.
Ask which elements the supplier can replace without your agreement and what notice you receive. Two products with different names may still rely on the same underlying service. Buying a second subscription does not automatically create an independent fallback.
The NCSC's cloud-security guidance recommends understanding how third-party suppliers access data and how a provider manages security across its supply chain. That guidance concerns security; commercial continuity also requires examining contracts, capacity and support. [3]
Keep the exercise proportionate. A tool used occasionally to suggest internal wording deserves less attention than a system through which every customer order passes.
Understand the economics of your own contract
Ask what the quoted price includes, which usage limits apply and what happens when the business grows. A low introductory charge tells you little about the cost of running the workflow at full volume.
Discuss price-change provisions, minimum commitments and support arrangements. Where a service depends on a named model, establish what happens if that model is withdrawn or its terms change. Request evidence for important assurances rather than treating a funding announcement as proof of long-term service capacity.
There is no need to diagnose a supplier's balance sheet from newspaper coverage. The objective is to understand your exposure and obtain answers that can be reflected in the agreement.
Rehearse a manageable interruption
Choose a realistic scenario: the service is unavailable for a working day, a model is retired or a key integration stops functioning. Identify which customers would be affected and what work could continue manually.
Check that authorised staff can retrieve the records and configuration they would need. An export is useful only if the business can read it, protect it and use it elsewhere. Test a small, non-sensitive example rather than assuming that an export button provides a complete recovery plan.
Assign someone to keep the plan current as the workflow changes. Include the time and cost of switching in the purchasing decision, alongside the productivity gains.
AI can support growth without every experiment becoming a permanent dependency. The stronger commercial position is to know which services matter, what sits behind them and what the business would do if the arrangement changed.
Sources & further reading
- Federal Reserve / Michael S. Barr: Some Potential Financial Stability Implications of Artificial Intelligence ↗Published 16 November 2026
- Reuters / Investing.com: Fed’s Barr warns on AI risks to financial system ↗Published 16 November 2026
- National Cyber Security Centre: Cloud security principles: Principle 8, Supply chain security (reviewed 7 June 2023) ↗Published 17 November 2018
Let’s talk
Chris McDonald’s perspective
I would ask one practical question before making an AI tool essential to customer delivery: how would we keep working if it changed tomorrow? A clear answer helps a business adopt useful technology with a better understanding of the commitment it is making.