Data Center Decommissioning in Virginia

Northern Virginia is the densest concentration of data center load in the world, and it is also, quietly, a place where a great deal of legacy capacity is being retired. Enterprises that built their own halls in Ashburn and Sterling a decade ago are migrating into cloud or into modern colocation. Providers are consolidating older buildings into newer ones. Federal contractors are collapsing multiple legacy facilities into a single certified environment.

Every one of those migrations produces an energy decommissioning, and data centers are the building type where decommissioning most reliably fails to deliver the projected savings. The reason is that almost none of a data center's cost structure is variable with the thing being removed. The racks go. The infrastructure that served them, the capacity reserved to feed it, and the commitments signed to guarantee both remain exactly where they were.

Emptying a Hall Makes It Less Efficient, Not More

The first-order effect nobody models correctly: a partially emptied data hall runs at worse efficiency than a full one.

Mechanical and electrical infrastructure in a data center has a large fixed component. Chillers and pumps have minimum turndown. Computer room air handlers run to maintain the room, not the rack. UPS systems have their own conversion losses and run less efficiently at low load factor than at design load. Lighting, controls and building systems are indifferent to how many servers are present.

Pull half the IT load out of a hall and you have not halved the facility load — you have moved the site to a worse point on its efficiency curve, where a growing share of the power drawn is overhead. Power usage effectiveness, the ratio most operators track, gets visibly worse during a migration, and it gets worse for a reason that reflects physics rather than management.

This has a direct operational implication: consolidate before you migrate. Emptying racks evenly across rows and rooms is the natural way to run a migration and the most expensive. Concentrating remaining load into the fewest rows, rooms and electrical distribution paths lets whole cooling zones be secured and produces a step reduction rather than a gradual worsening. It requires planning the migration order around the mechanical topology rather than around application dependencies alone, which is a conversation between facilities and infrastructure teams that frequently does not happen.

Contract Demand Is a Commitment, Not a Measurement

Large loads in Virginia are generally served under tariff arrangements that involve a contracted or reserved level of capacity. The utility plans, builds and maintains distribution and, at scale, transmission-level facilities to serve that level. The charge reflects that commitment, and it is why a large service can carry a substantial monthly bill against low consumption.

Three consequences matter when decommissioning:

• Removing load does not reduce the reservation. The reservation is contractual. It changes when you ask for it to change, in writing, subject to the tariff's conditions — which commonly include notice periods, minimum terms, or floors tied to historical maximum demand.
• The reduction is prospective. Months between the migration and the request are billed at the old level, and nothing was billed incorrectly, so there is nothing to recover.
• Reducing it may be irreversible in practice. Releasing capacity in a constrained region is not necessarily a decision you can reverse on demand. In parts of Northern Virginia, capacity for large new loads has been genuinely scarce and interconnection timelines long. An enterprise that releases a large reservation and later wants it back may find itself in a queue behind hyperscale demand. That is a strategic decision, not a facilities decision, and it should be made by someone who understands the company's five-year compute plans.

The last point is the one that distinguishes data center decommissioning from every other kind. Elsewhere, releasing reserved capacity is straightforwardly good. Here, the reservation may be the scarcest asset the company holds at that site — and worth evaluating for its value in a sale or sublease of the building rather than simply surrendered.

The Tags Were Set While the Hall Was Full

Virginia sits in PJM, where capacity and network transmission charges are allocated to accounts through load tags derived from usage during prior-summer peak hours and applied across a delivery year running June through May.

Data centers carry structurally high tags because they have flat load. A building that can shed load on a hot afternoon contributes less to its tag than its average draw; a data center contributes essentially its full draw to every peak hour, because that is what it is designed to do. A hall that migrates out in the fall therefore carries a tag measured at full population into the following spring, with no ability to shed it and no operational lever to pull.

The implication for the migration plan is a scheduling one. A migration that completes before a summer avoids setting a full-population tag for the following delivery year. A migration that completes just after a summer carries that summer's tag for twelve more months. Where the migration date has any flexibility at all — and enterprise migrations usually have several months of it — that is a meaningful number and it is entirely knowable in advance. The broader mechanics are in understanding capacity charges and why PJM capacity charges are rising.

If You Are Leaving Colocation Rather Than Your Own Building

Most enterprise migrations out of Northern Virginia are exits from colocation, not from owned facilities, and colocation agreements have their own decommissioning economics that have nothing to do with utility tariffs.

Colocation is sold on committed capacity in kilowatts. Vacating the cabinets does not reduce the commitment, and the minimum charge continues for the term. The provisions worth locating before the migration begins:

• The commitment and its step-down rights. Some agreements allow the committed draw to be reduced at defined intervals with notice. Many do not. This is the single most valuable clause to have negotiated at signing and the least likely to have been read.
• Portability across the provider's portfolio. If you are consolidating into another facility operated by the same provider, the commitment can often be moved rather than terminated. Providers are considerably more flexible about retaining a customer than about releasing one.
• The metered power component. Where power is billed on metered draw above the commitment, the migration reduces this immediately — which is the part of the bill that does fall, and is often mistaken for the whole bill falling.
• Cross-connect and remote hands. Recurring charges for cross-connects survive the removal of the equipment they connected until they are explicitly canceled. This is the colocation equivalent of a phantom account and it is extremely common.
• Decommissioning and restoration obligations. Cage removal, cabinet disposal and restoration of the white space to the provider's standard are usually the customer's cost.

What Stays Energized in the Building You Are Leaving

For an owned or leased facility being wound down, the accounts and loads that survive the decommissioning are predictable:

• The UPS and battery plant. Kept energized to preserve battery health while the disposition of the equipment is decided, sometimes for a year.
• Generator block heaters and controls. Standby generation is maintained on a schedule that continues until the units are sold or removed, and block heaters draw continuously.
• Fire suppression, security and building management systems. Legitimately required while the building is held.
• Minimum conditioning. A held white space usually maintains humidity control to protect the remaining infrastructure.
• Fiber and carrier services. Not electricity, but the same failure mode: circuits ordered years ago, billed monthly, and never canceled because the person who ordered them has left.

Each of these is defensible individually. Collectively, they are why an empty data center in Loudoun County can carry a monthly cost that surprises everyone when it appears in a variance report six months after the migration was declared complete.

The Sequence That Works

Establish the disposition of the building first, because it determines everything else — held for re-tenant, sold, or returned to a landlord. Model the tag and contract demand cost for a migration completing before versus after the coming summer, and let that inform the schedule. Consolidate remaining load into the minimum mechanical footprint rather than emptying evenly. Request the contract demand reduction in writing, having first decided deliberately whether releasing that capacity is something the company might regret. Cancel cross-connects and circuits explicitly. And ninety days after the last rack ships, reconcile the payables file against the facility list, because something will still be billing.