AI Deals Are Redefining What Counts as Capacity

Capacity now means more than a utility allocation or a planned substation. For AI infrastructure deals, buyers are starting to judge whether power, capital, operating terms, and a credible delivery path are all tied together. Operators that still market a single MW number are going to look incomplete.

That shift is visible in the kinds of agreements being announced. The biggest projects are pairing campuses with financing support, dedicated generation, flexible-load arrangements, and long-term customer commitments. Smaller colo deals are not exempt; they simply need to translate the same logic into a more practical proof package.

Why is a power number no longer enough?

For years, a capacity claim could be reasonably simple: a site had a utility service agreement, a certain amount of sellable power, and a target ready-for-service date. Buyers would still test those claims, of course, but the basic unit of comparison was the megawatt.

AI demand has made that shorthand less reliable. A buyer needs to know whether the power can serve its intended load profile, what happens during a grid shortfall, who pays for related upgrades, whether the facility can physically deploy the equipment, and whether the project has enough financial backing to get through construction. A large number without those answers is now just an early-stage lead.

The Ohio arrangement between NVIDIA, SB Energy, and OpenAI is an unusually clear example. NVIDIA said it would provide credit support for land, power, and shell buildout for an initial 4.25 IT GW at the PORTS-Pike campus, with an option covering the remaining capacity; OpenAI is slated to lease 8 IT GW over 20 years (NVIDIA Newsroom). The point is not that every operator needs a deal of that scale. It is that the announced capacity is inseparable from how it will be financed, built, and occupied.

In practice, buyers will increasingly ask for the same connection at a 5 MW or 50 MW requirement: “What specifically makes this capacity real?”

What are buyers really trying to de-risk?

They are not usually seeking a perfect forecast of the future. They are trying to avoid a painful failure between signed paper and commissioned cabinets.

That failure can show up in several places:

• The utility has identified service, but transmission work or a substation upgrade remains unresolved.
• The operator has a construction schedule but has not secured the equipment, capital, or site approvals needed to hold it.
• The buyer can take initial load, but the promised expansion tranche depends on conditions nobody has explained.
• The grid can serve normal operation, yet a curtailment or demand-response obligation changes the customer’s operating assumptions.

PJM’s proposed framework for new large loads makes the last point harder to treat as a footnote. The framework would establish Interim Resource Adequacy Service for new large loads that do not bring or secure new supply, and in scarcity events it would call for reducing those loads before traditional residential customers. PJM says data centers account for 30 GW of its projected 32 GW of load growth from 2024 through 2030 (PJM Interconnection).

Whether a particular customer is exposed to that framework is a diligence question, not a marketing slogan. But the broader signal matters: load flexibility, resource commitments, and contractual operating rights are becoming part of the capacity conversation.

How should operators describe capacity now?

I'd argue for retiring the phrase “available power” unless the next sentence explains its status. It is too easy for a prospect to hear “available” as “contract-ready and deliverable on my schedule,” while the sales team means “identified in a utility discussion.” That gap is where trust gets lost.

Build a capacity statement with distinct layers instead:

Serviceable now. State what is live, sellable, and physically deployable. Include the actual handoff point and any density or cooling constraint that affects the buyer’s design.

Contracted future service. Explain what agreement supports it, what construction must occur, and which milestone changes it from planned to usable. Do not bury dependencies in an appendix.

Expansion under evaluation. This can still be commercially useful, particularly for a customer with a long runway, but label it as pipeline rather than inventory. AIB Data Centers recently made this distinction well: it reported a 65 MW electric-service agreement for its existing CLT-01 facility while separately describing prospective capacity across sites still under evaluation and not under definitive development, lease, or purchase agreements (GlobeNewswire).

That discipline does not weaken a pitch. It gives a serious buyer a basis for deciding which portion of your offer can support a current procurement and which portion belongs in a future option discussion.

Does behind-the-meter power change the sales story?

It changes it, but it does not remove the need for diligence. The emerging model is less “the data center has solved power” and more “the data center has added controllable resources and clearer options around power.”

Qcells and Microsoft, for example, said they are exploring development of new generation and flexible energy resources alongside Microsoft data centers, with potential delivery to Microsoft directly or through the local utility. Their work also includes exploring virtual power plants that aggregate batteries for peak-grid support (Qcells North America).

For an operator, the useful lesson is not to promise an imitation of a hyperscaler strategy. It is to get specific about the role of on-site or adjacent resources. Are they intended to support resiliency? Reduce peak exposure? Meet an interconnection condition? Serve as a bridge until utility capacity arrives? Each answer creates a different commercial and operational commitment.

Calling every battery, generator, or flexible-load arrangement “dedicated power” will invite a tough technical review. Explain the operating model instead. The right buyer will appreciate the precision.

What does this mean for mid-market facilities?

The mid-market opportunity is not to compete on the largest theoretical campus. It is to be easier to underwrite.

A regional operator can often offer something a giant plan cannot: a defined deployment phase, a known building, direct access to the people handling utility coordination, and a plain explanation of what must happen before turn-up. Those are valuable advantages when an enterprise team is trying to put a deployment into production rather than make a strategic land bet.

Take a smaller AI colo transaction: QumulusAI announced a seven-year agreement for up to 3.75 MW in metro Atlanta, plus a right of first offer for adjacent capacity, with readiness tied to an incumbent tenant’s transition (Business Wire). What makes that announcement credible is not merely the initial capacity. It identifies the term, the expansion mechanism, and the dependency affecting readiness.

Your sales materials should do the same. A prospect should be able to see the initial block, the next block, the condition attached to each, and the decision deadline for reserving it. That is far more useful than a site map with a large “future capacity” label.

Common questions

Should we stop marketing future capacity?

No. Future capacity matters to buyers with phased deployments and growth plans. Market it as future capacity, with the milestones, dependencies, and reservation path that make it meaningful rather than presenting it as live inventory.

How much detail should sales share about grid constraints?

Share enough for the buyer to understand the commercial consequence: timing, cost responsibility, curtailment exposure, and the party accountable for each next step. Technical diligence can go deeper under NDA, but vague answers in an early call usually end the conversation.

Will buyers expect their own generation commitments?

Not every buyer will, and it depends on market and load profile. Still, the growing interest in bringing new supply or flexible resources alongside load means sophisticated buyers may ask whether your site can support, accommodate, or contract around such a model.

What is the best proof of capacity in a first meeting?

Start with a one-page capacity schedule that separates live service, contracted delivery, and conditional expansion. Pair it with the relevant utility status, construction milestone, and a named technical owner who can answer follow-up questions.

Where this leaves you

The market is moving from capacity claims to capacity evidence. Treat power, delivery, operating constraints, and expansion rights as one commercial story, and your team will have a much more credible conversation with serious AI buyers. GridReach helps data center and energy companies turn expertise like this into qualified pipeline.