The missing balance question of the AI ​​boom: What is actually verified?

Artificial intelligence infrastructure commitments are increasing faster than evidence that contracted capacity has sustainable economic value.

I spent 13 years on the floor of the New York Stock Exchange. One lesson stayed with me long after the trading jackets disappeared: a transaction could be executed, documented, and perfectly valid, while the assumptions that made it attractive proved less tenable than expected.

This distinction is important in the boom in artificial intelligence infrastructure. We can assume that AI will transform productivity and still question whether every financing, data center expansion, or long-term capacity agreement is supported by evidence that can outlast the life of the investment.

The industry makes commitments across different clocks. Compute contracts can run for years. Debts can come due in decades.

Servers have much shorter technology cycles. Energy infrastructure, transmission upgrades and buildings can outlast multiple generations of the equipment they were built to support. Meanwhile, enterprise customers are still learning what they will use, how much they will use it, and what they will pay.

None of this proves that the investments are unsafe. It raises a question that goes beyond revenue forecasts and capital budgets: What exactly was being tested at the moment an institution decides to rely on a promise?

Contracted demand is not the same as productive use

An agreement to purchase computing power can make economic sense. It can support financing and demonstrate customer commitment. However, it alone does not ensure that the underlying workloads will remain commercially productive to the extent contracted or that the customer will renew on similar terms as technology and pricing change.

A serious examination must separate the existence of the contract from the economic assumptions underlying its duration, usage profile, counterparty capacity and renewal expectations. Contract enforceability and economic durability are related; they are not interchangeable.

Another milestone is the physically deliverable capacity

A data center campus can be announced, funded, and partially constructed before all necessary power, cooling, network, and equipment dependencies are operational. A power purchase agreement alone is not proof of the ability to deliver on the exact commissioning date. GPU orders are not evidence that a facility can operate the ordered cluster within the planned duty cycle.

For infrastructure planning, I use a simple order: requested load, proven load, deliverable load, activated load, used load and sustained demand. These aren’t six fancy names for the same megawatt. They describe different claims, each of which requires different evidence.

This distinction is already relevant for decisions about public networks. The Persistence Analytics Group provided comments at the MISO Large Load Working Group meetings on load forecasting, application and study processes, paired resource studies, and cost transparency. The perennial question is how planners maintain alignment when a proposed project’s capacity, schedule, or other essential conditions change after a previous study or commitment.

A guarantee allocates risk; it doesn’t eliminate it

Financial engineering can offer useful protection. Minimum payments, parent guarantees, collateral requirements, termination provisions and residual value commitments can postpone or mitigate losses. They should be evaluated based on their actual conditions and creditworthiness.

However, a contractual remedy is not the same as the continued usefulness of the underlying asset. A guarantee can support recovery from a counterparty while leaving the investor dependent on that counterparty’s resources, enforceability, timing and remaining value following a disruption. These are separate assumptions that deserve separate evidence.

This is particularly important when seemingly disparate risks are based on a small group of interconnected customers, infrastructure providers, funding sources or technology assumptions. Diversifying contracts does not necessarily lead to diversifying their underlying dependencies.

The missing control is a revalidation

The initial review typically asks whether a project is eligible for funding upon approval. A more sustainable approach asks what needs to remain true afterward—and what changes require a re-examination of the original case before providing additional money.

Imagine a campus whose customer delays ramp-up, whose power availability is postponed until a later year, or whose equipment changes significantly. Each event can be managed on its own. However, if the original funding or infrastructure decision was based on the previous configuration, someone must decide whether the updated evidence still supports the next commitment.

This is a decision point, not a prediction about whether AI will be successful. It requires a documented chain: the representation relied upon; the assumption embedded therein; the evidence supporting this assumption; the dependencies that must exist; the party bearing downside risk; and the material change that would require reassessment.

Different specialists own different pieces. Engineers assess technical delivery capability. Lawyers assess contractual rights. Accountants and finance teams evaluate financial statements and cash flows. Independent decision assurance asks whether these records, taken together, actually justify the proposal on which the capital allocator relies – and whether known gaps have been made explicit.

The question CFOs should put on the table

The debate about productive use is essential. There is also the question of who will pay for AI intelligence and at what price. But before approving the next investment, a CFO or investment committee may ask an even more immediate question: Which assumptions are verified facts, which are binding commitments, which are conditional forecasts, and which need to be reconfirmed because something material has changed?

A sector can produce transformative technologies and produce disappointing returns for some of its assets. We have already seen this difference in markets. The lesson is not to stop investing. The point is to no longer treat execution, delivery, use and enduring economic value as one and the same event.

On the trading floor, confirming a trade was the beginning of understanding the position, not the end. AI infrastructure deserves the same discipline.

Neil P. Osnato is the founder of Persistence Analytics Group LLC, an independent Decision Assurance™ and Infrastructure Assumption Verification™ company. He previously worked at the New York Stock Exchange for 13 years. The focus of his work is on documentary evidence supporting resulting infrastructure and financial decisions. He wrote this article for SiliconANGLE.

Selected public background: MISO Large Load Working Group stakeholder feedback (LLWG-2026-7 and LLWG-2026-8), September 2026; The Bond Buyer, “Data Center Decision,” October 28, 2026.

Image: The Digital Artist/Pixabay