Facility Decommissioning, Repowering & Site Remediation

Chapter 14.9 retired the silicon: the servers were sanitized, the GPUs went into the ITAD channel and the secondary market, and the white space emptied. This chapter retires the building: the shell, the substation, the generator yard, the fuel farm, the BESS, the cooling plant and the dirt underneath all of it. That is a different lifecycle stage, with a different clock, owner and liability surface. In the contested bear case, a frontier accelerator is economically obsolete in 2–3 years. A powered shell, a substation and a permitted interconnection run on a 20–40 year clock. The IT inside turns over six or eight times before the building reaches a decision point, and at that point retained rights and physical suitability decide whether keeping the shell beats conversion or demolition.

The central choice, repower vs demolish vs adaptive reuse, comes first, along with why a power-bound market tilts it toward keeping the asset. Next is the physical work: pulling generators, closing fuel tanks, de-energizing and recycling the BESS, and disposing of the coolants, glycol, dielectric fluid and PFAS that liquid cooling put into the building. Then come two instruments: the decommissioning bond, which the 2025–2026 model-ordinance wave is pushing into local entitlement conditions, and the Phase I/II ESA exit with its regulatory closure letter. The chapter ends on the restoration obligations buried in leases, host agreements and permits, and on the design choices made decades earlier that make all of this cheap or ruinous.

End-of-life is a choice among three terminal states, each with a different cost structure and a different downstream liability. Make the choice at the design-basis stage and revisit it at every refresh, because the cheapest decommissioning is the one you engineered for twenty years earlier.

Repower (retain-and-rebuild). The shell, slab, substation, switchgear and interconnection stay. The IT, and usually the cooling and UPS plant, are gutted and replaced for the next density generation. Repowering can shorten time to capacity when retained grid rights, approvals and the physical retrofit beat the alternatives. DOE's Section 1706 loans, first branded Energy Infrastructure Reinvestment (under the IRA) and rebranded Energy Dominance Financing in the July 2025 OBBBA, fund the retooling, repowering or repurposing of retired energy infrastructure. For a coal or gas site the argument is the same: the grid connection is what is scarce, and reusing it is faster than waiting in the queue. But a retired plant's interconnection agreement is a right to inject power, not to draw it, so a data center on the site still needs the transmission owner's load study; what carries over is the switchyard, the lines and the rights-of-way. The catch is the density-ramp ceiling. A legacy air-cooled shell cannot absorb a GB300 NVL72 (132 kW nominal rack TDP in HPE's specification, up to 142 kW in NVIDIA's reference architecture, 155 kW peak at Lenovo), or a Vera Rubin NVL72 at the 330 kW cabinet TDP that NVIDIA's DSX reference design plans for. The ~600 kW Rubin Ultra / Kyber roadmap planning point is higher still. Absorbing any of them means reworking the floor structure, re-running the power chain and plumbing for liquid the building never had. Repowering is cheap only if the irreversible substrate was provisioned for the ramp.

Retained rights are a contract, not a property right. Before you assume the megawatts survive a repower, read three terms in the electric service agreement or large-load tariff. The first is the minimum-demand charge you keep paying while the white space sits empty. AEP Ohio's Schedule DCT, which the Ohio PUC approved in July 2025, sets that minimum at no more than 85% of contract capacity and bills it even while service is curtailed or suspended (Chapter 2.4 works the numbers). Exiting early costs 36 months of those minimum charges. The second is any clause that lets the utility reclaim capacity you leave unused. The third is whether the added load needs a fresh load study; any repower that lifts demand above the contracted capacity does. Existing AEP Ohio agreements are grandfathered only until the site expands by more than 25 MW, so a repower that adds load can also lose that protection.

Demolish-and-restore. The building comes down and the site returns to a baseline that a lease, permit or host agreement defines. The budget combines structural demolition with environmental closeout: fuel-tank closure, scoped soil investigation and any required remediation. Phase II sampling tests specified contaminants at specified locations; it cannot prove all the dirt clean. Demolition is the right call when retrofit cost or schedule exceeds the retained assets’ value, when the interconnection is stranded or de-rated, when the market has no demand, or when a restoration clause makes holding the asset more expensive than surrendering it.

Adaptive reuse. Either the shell takes on another use, or the conversion runs the other way, into a data center: a fab, warehouse, brewery, printing plant or retail anchor becomes a data center that inherits the building's power, water and zoning. Adaptive reuse is the inbound mirror of repowering: the logic that says 'keep the energized shell' also says 'buy the energized shell someone else is abandoning.' It unlocks dense interconnection near population centers that greenfield siting cannot reach (DCD, 2025–2026).

If the decision is demolish, or a deep repower that strips the power and cooling plant, the physical work is governed by what is hazardous, what is recoverable and what is regulated. Three subsystems dominate the scope and schedule: the generator and fuel farm, the energy-storage system and the cooling fluids. Each carries its own permit, its own waste manifest and its own way of going wrong.

Generators and the fuel farm. A standby fleet runs on diesel, and diesel means tanks. Day tanks, bulk above-ground storage tanks (ASTs) and any underground storage tanks (USTs) are regulated closure events. The fuel is pumped out and recovered or disposed of, lines are purged, the tank is cleaned and either removed or closed in place under the governing UST/AST program, and the soil beneath and around it is sampled. A historical leak turns a routine tank pull into a contaminated-soil excavation, a common and open-ended addition to a decommissioning budget. The generators themselves are valuable recoverable assets, since gensets, transfer switches and switchgear have an established secondary market. The catalytic/SCR aftertreatment and any residual fuel, however, must be handled as regulated streams. Where the site ran gas turbines or fuel cells as primary or bridge power (→ Chapter 3.5), the gas interconnection and process-safety isolation become part of the closeout.

The BESS and the UPS battery estate. The two chemistries have different end-of-life paths. Legacy VRLA/lead-acid UPS strings are the easy case. Lead batteries are recycled through a mature closed-loop chain at a ~98% rate (BCI, 2020–2024, all lead batteries), so the strings are a recoverable commodity. Lithium, increasingly LFP for facility-scale BESS doing ride-through, transient smoothing and demand response, as in NVIDIA's BESS reference designs, is the hard case. A multi-MWh lithium installation must be safely de-energized, discharged to a transport state of charge and shipped as a Class 9 hazardous material to a qualified recycler. Damaged or thermally abused cells carry a real fire risk through the whole removal and transport chain. Recycling economics are improving fast but are not yet the lead-acid commodity loop, and the regulatory framing (UN 38.3 transport, state battery-stewardship laws) is tightening. Budget BESS removal as a hazmat operation, not a scrap pull.

Beyond PFAS, a liquid-cooled AI facility holds a real inventory of working fluids that must be characterized and disposed of at end-of-life. The technology-cooling loop runs propylene-glycol/water blends (PG-25-class) with biocides and corrosion inhibitors. The facility-water loop may run glycol for freeze protection. Single-phase DLC uses dielectric or water-glycol coolants. Refrigerant-bearing chillers hold regulated refrigerants subject to recovery rules, and transformers and some switchgear hold dielectric oils that must be tested for PCBs before disposal. None of these is dramatic alone, but together they add up to thousands of gallons of inhibited, biocided fluid that cannot go to a storm or sanitary system. Each stream is sampled, characterized against the disposal facility's acceptance criteria, manifested and tracked. All of it is far cheaper if the fluid systems were segregated and documented at construction. A building with mixed, undocumented loops forces you to characterize everything from scratch, at hazmat-lab prices, on the demolition critical path.

The financial spine of facility end-of-life is an obligation specific to each project. List what binds it: the enacted ordinance, any entitlement condition, the lease or restoration covenant, the required instrument, the sizing formula, the re-estimation interval, the release test and the effective date. York County's model ordinance (August 2025, revised April 2026) shows how suggested language can turn the decommissioning bond into a municipal entitlement condition. It sets security at 110% of the estimated decommissioning cost minus salvage value, with the municipality reserving the right to review once every five years, and it does not impose automatic escalation. York's security covers removing the equipment specific to a data center (cabling, electronics, cooling, power generation, batteries) and leaving a building and property that can be reused. It is not sized for demolition to grade, and the owner posts it before the certificate of occupancy. If a lease or host agreement requires demolition or full site remediation, that obligation needs separate security. The model’s wording does not say whether the 110% applies before or after salvage is subtracted, so read the adopted text for that, for what counts as salvage and for the release test.

The developer proposes the form of security, typically a surety bond, a letter of credit or a cash escrow, and the municipality must accept it. The form is itself a financial decision. A surety bond is cheapest on the balance sheet but underwritten against tightly defined triggers and long-duration assumptions. A letter of credit ties up borrowing capacity as collateral. A cash escrow is the cleanest for the municipality and the most punishing for the developer's working capital (Taft Law).

The consequences run straight into project finance. A bond sized at construction can change when the binding instrument permits re-estimation, and a five-year review can raise or lower the amount, so lenders price the actual triggers and collateral requirements. The salvage-value offset is where engineering meets finance. A facility designed for clean disassembly (modular, documented, with recoverable gensets, switchgear and copper) carries a higher salvage credit and therefore a smaller net bond, while a monolithic, undocumented build maximizes the bonded amount. Drawdown mechanics matter too. The adopted security instrument identifies the approving authority, completion evidence and any staged release, and environmental closeout and financial-security release need separate evidence under their respective instruments. Neither a Phase I nor a generic closure letter automatically releases the bond. → community/host-agreement framing in Chapter 3.11; the build-vs-lease optionality this protects in Chapter 1.8.

The environmental closeout follows a standardized due-diligence ladder, and knowing it is the difference between cleanly severing liability and inheriting it. A Phase I ESA (per ASTM E1527-21, the standard EPA has recognized for All Appropriate Inquiries since February 2023) is a non-intrusive study of records, interviews and the site itself. It identifies recognized environmental conditions (RECs), meaning evidence of a release or threatened release. A compliant Phase I satisfies the All Appropriate Inquiries threshold that the CERCLA bona fide prospective purchaser and innocent landowner defenses require. Those defenses also demand continuing obligations after closing (reasonable steps on discovered releases, cooperation with regulators, compliance with institutional controls), so the Phase I starts the protection rather than completing it. If the Phase I flags a REC (a former UST, a stained slab, a documented spill), a Phase II ESA follows: intrusive soil, soil-vapor and groundwater sampling to confirm or refute contamination. Phase II scope follows the suspected release, media, sampling design and laboratory program. A clean result does not imply a cheap investigation, and detected contamination does not imply a fixed cost tier.

For a data center, the predictable RECs are the diesel fuel farm (the dominant one: decades of tanks, lines and refueling), transformer dielectric oils (PCB testing), the BESS footprint and any PFAS-bearing fluid history. Since July 2024, PFOA and PFOS have been CERCLA hazardous substances, so under ASTM E1527-21 a Phase I must evaluate them. On 18 August 2026 the D.C. Circuit upheld the designation (Chamber of Commerce v. EPA), and the window to petition the Supreme Court runs to mid-November 2026. At a data center the usual source is legacy AFFF from fire protection or the fuel farm, not modern fluorinated coolants. A Phase I that skips PFOA and PFOS weakens the purchaser defenses it is meant to secure. A release of one pound or more of either within 24 hours must be reported to the National Response Center (EPA).

The exit you are buying is a regulator's closure: a 'No Further Action' letter confirming the site meets the cleanup standard for its intended use. It is the strongest closeout available, but not an absolute release. A state closure does not automatically bind federal authorities, most letters carry reopeners, and unknown conditions sit outside them. The asymmetry is scope: a Phase I documents recognized environmental conditions without intrusive sampling, while Phase II and any remediation have a site-specific extent and can add years. A baseline Phase I at acquisition informs that history but cannot prove that every subsurface condition predates ownership. Preserve acquisition history, release evidence and continuing-obligation records, because the purchaser defenses hold only while their statutory conditions are met.

The costliest mistake in facility end-of-life is discovering at decommissioning that 'restored' means something stricter than you assumed. The restoration baseline is not one document. It is the combined applicable obligations in four instrument groups, and they rarely agree. The ground lease or surrender clause dictates the condition in which a leased site or building must be returned: sometimes 'broom-clean,' sometimes 'remove all improvements and restore to grade,' a clause whose cost can dwarf the salvage value. The host-community agreement / CBA / PILOT increasingly carries its own restoration and bonding language as the price of the original rezoning (→ Chapter 3.11). The conditional-use / special-exception permit may condition the entitlement on a decommissioning plan and a return-to-baseline obligation. The environmental permits (the air permit covering the generator fleet, the NPDES/water-discharge permit and the stormwater permit) each require formal termination, and a permit you forget to close out keeps accruing obligations against an asset that no longer exists.

The procedure is mechanical: inventory every binding instrument at acquisition and again before decommissioning; satisfy each applicable obligation and resolve conflicts with the relevant authority before work. A team that restores to the lease's broom-clean standard and then discovers the host agreement required full demolition to grade has not finished the job. It has created a default. The bond will not release, the permits will not terminate, and the contingent liability stays on the books. Close each obligation with the evidence its instrument requires. Record remediation closure, waste receipts and lease restoration separately, then obtain any drawdown or final security release under that instrument’s own test.