A Power Deal Is Not IT Capacity

A long-term power purchase agreement can be strategically important without proving that a single data center has secured an equivalent amount of deliverable IT capacity. That distinction matters more as hyperscalers sign multi-gigawatt energy contracts and grid operators redesign the rules for large loads.

On 6 October 2026, Google and Constellation announced a collaboration intended to add 890 MW of nuclear capacity to the PJM system. Reporting on the broader arrangement described 3,590 MW of contracted supply across nuclear uprates and other PJM resources. Those are important power-market facts. They are not, by themselves, a 3,590 MW data-center campus, a 3,590 MW grid connection at one address, or 3,590 MW of IT load.

For investors, developers and occupiers, the practical question is not simply “how much power?” It is: what exactly has been contracted, where can it be delivered, when does it become available, and what portion can support critical IT load after electrical and cooling losses?

Five capacities that must stay separate

A disciplined market model keeps at least five concepts distinct.

Generation capacity is the nameplate or incremental output of a power plant. The Google announcement describes 890 MW of additional nuclear capacity for the PJM grid. Constellation’s release uses the same basis.

Contracted energy or power is the commercial volume covered by an agreement. It can span several generators, nodes and delivery periods. Reuters reported that the wider arrangement includes 890 MW under a 20-year nuclear agreement and 2,700 MW under a separate 15-year agreement in PJM.

Grid connection capacity is the amount a particular site is authorized and physically able to import or export at a defined point. It depends on studies, network upgrades, protection schemes, milestones and sometimes curtailment conditions. PowerlandMap’s guide to grid-connection deliverability explains why an application, queue position and executable connection agreement are different evidence grades.

Facility electrical capacity is the power available to the campus or building after the connection architecture, substations and redundancy design are considered. It can be expressed as utility feed, transformer capacity, contracted capacity or maximum demand; those are not always interchangeable.

IT capacity is the critical load available to servers and network equipment. It is lower than facility input because cooling, power conversion and other infrastructure consume energy. Power usage effectiveness can help bridge the concepts, but only when the measurement boundary and operating assumptions are known.

The PowerlandMap coverage model therefore stores sourced IT MW separately from campus, electrical, generation and future potential.

Why a portfolio agreement does not locate capacity

A portfolio-level agreement may support growth across many regions and facilities. It can hedge exposure, finance additional generation, improve system reliability or satisfy clean-energy targets. It does not necessarily identify the benefiting data centers.

Location matters because electricity is delivered through a constrained network. A megawatt added in one zone does not automatically become a megawatt available at every substation. Congestion, transmission outages, interconnection upgrades and local distribution constraints can all change the usable result.

PJM’s proposed framework asks new large-load customers to build, bring or buy new supply and bear its full cost. The PJM proposal is a commercial and regulatory pathway; it is not a guarantee that every proposed load will obtain firm service on its preferred date.

This is why the PowerlandMap constructability workflow evaluates the site, power path, permits and delivery schedule together. A strong energy procurement strategy improves the case, but it does not replace site-specific diligence.

The evidence ladder for a credible capacity claim

Analysts can avoid most overstatement by using a simple evidence ladder.

At the lowest level is a corporate ambition: a company plans to procure or enable a certain volume of power. Next comes an executed commercial agreement with identified counterparties and terms. Above that is a documented connection pathway for a named site, followed by permits and funded network works. The strongest evidence is an energized facility with a verified measurement boundary and operational date.

Each rung answers a different question. An executed PPA proves more than a memorandum of understanding, but less than an energized substation. A connection agreement proves more than a queue application, but may still depend on upgrades or milestone compliance. A building announcement can show development intent without proving critical IT load.

The same discipline applies to behind-the-meter projects. On-site generation can reduce exposure to grid constraints, yet fuel supply, emissions permits, reliability standards, islanding capability and backup architecture still determine deliverability.

Flexible load changes the contract, not the physics

Large-load regulation is increasingly considering flexibility. The Federal Energy Regulatory Commission has directed regional grid operators to justify or reform how they connect data centers and other large users. Flexible service, curtailment and workload shifting can accelerate some connections or reduce required upgrades.

But flexibility changes service conditions. It should be recorded as a contractual and operational attribute, not translated into firm capacity. A 100 MW connection subject to curtailment is different from 100 MW of firm service. A workload that can move between regions may have portfolio resilience without providing continuous capacity at any single site.

This is also why grid-interactive data centers need separate fields for baseline load, interruptible load, on-site storage, generation and response obligations.

What investors and occupiers should ask

A capacity claim becomes decision-useful when it answers a short set of questions:

1. Is the number generation MW, contracted MW, facility electrical MW or IT MW? 2. Is it attached to one site, a regional portfolio or the whole grid? 3. Is the service firm, interruptible, phased or conditional? 4. Which substations, voltage levels and upgrades are involved? 5. What are the target energization and ready-for-service dates? 6. Which milestones could terminate or resize the right? 7. Has the IT conversion used a disclosed PUE and measurement boundary? 8. Is the capacity live, under construction, permitted or merely planned?

These questions belong in underwriting, commercial diligence and market intelligence. They help separate a strong strategic signal from a site-level claim.

PowerlandMap’s Market Intelligence workspace keeps those evidence layers visible, while its product overview and use cases show how they support developers, investors and occupiers.

The market signal is still real

Keeping the categories separate does not diminish the Google–Constellation agreement. It strengthens the interpretation. Adding 890 MW of nuclear output and contracting further PJM supply show that hyperscalers are moving upstream into generation, grid strategy and long-duration procurement.

The IEA’s 2026 analysis reinforces the structural pressure: data-center electricity demand continues to rise faster than the wider power system in many markets. That makes power procurement a competitive advantage.

The conclusion is precise. A major power agreement is evidence of capital commitment, procurement capability and strategic demand. It becomes evidence of deliverable data-center IT capacity only when the site, connection, timeline, service conditions and conversion basis are also proven.

For teams evaluating a project or portfolio, request access to compare sourced capacity, power-path evidence and delivery milestones without mixing generation, grid headroom, campus power and IT load.