Energy Supply Strategy: Grid PPA, BYOP & Co-Location

By 2026 the hard question is no longer "can I buy the chips?" but "can I energize them, and at what price, firmness and carbon profile?" Chapter 3.2 showed that the grid queue is the long pole: 3 to 7-plus years end to end, while PJM's generator queue now runs past 8 years from application to COD. Chapter 3.3 showed that power is a dominant operating expense and a cost lever you control. Those two facts collide in the energy supply portfolio: where your electrons come from, on what contract, at what price certainty, and against what regulatory and market risk. Few commitments in the project are harder to reverse. The contract tenor, the interconnection class and the on-site generation all outlive the GPU generation they were scoped to serve.

The portfolio has layers. First come the four supply structures and the fork between them. Next are PPA structures: physical versus virtual, fixed versus indexed, and the 24/7 carbon-free-energy matching debate that drives firming cost. Then power-market risk management, treated as a first-class engineering subject (nodal basis, VPPA shape and negative-settlement risk, congestion-revenue rights, heat-rate hedging), because an ERCOT-merchant exposure can swing a pro-forma further than any cooling optimization. Long-duration firm supply (nuclear restarts, uprates, SMRs, renewables-plus-storage firming) follows, along with the co-location rules that FERC's PJM orders of December 2025 and June 2026 reshaped. Last is the tenor-matching problem that ties a 17-year contract to a GPU whose economic life may be as short as the contested 2–3-year bear case. Most of these structures relocate risk rather than eliminate it, and the operator's job is to know which risk it has chosen to hold.

Every energy supply strategy blends four archetypes, and naming the dominant one is the first decision. Grid-only takes service under the host utility's large-load tariff and a signed agreement. Unit cost, firmness, studies, upgrades, interruption rights and the milestone schedule all depend on the jurisdiction and the project; there is no standard national load queue. In PJM, grid-only load should also expect a share of the one-time Reliability Backstop Procurement. Bids open 30 September 2026, for capacity commitments of up to 15 years at a weighted-average cap of $555/MW-day. PJM allocates the procurement's cost to load-serving entities, and whether they pass it on to large loads is up to the states (→ Chapter 3.3).

Grid + bridge generation energizes early on temporary or permanent on-site generation (typically gas turbines or reciprocating engines) and transitions to grid power as the interconnection completes. First power takes roughly 18–36 months when the units are reciprocating engines, aeroderivatives or turbine slots you already hold (→ Chapter 3.5 for lead times). The bridge buys speed at the cost of fuel, emissions permitting, and the risk that it becomes permanent. Behind-the-meter (BTM) island bypasses the grid entirely with dedicated on-site generation. It gives maximum speed and independence from the queue, but you now own the reliability, fuel-supply and firmness problems yourself. Hybrid / co-located sites the load next to a generator (gas, nuclear, or renewable-plus-storage), draws primary power from it, and contracts some grid service for backup or supplemental supply. It is the structure that FERC's PJM orders of December 2025 and June 2026 gave a regulatory home in PJM.

The four structures hedge different failure modes, so the 2026 default is to layer them rather than pick one: a co-located gas plant for first power, flexible grid service underneath it, a long-dated clean PPA for the carbon claim, and storage for ride-through. The behind-the-meter wave shows the point. Roughly 90 GW of BTM data-center generation has been announced cumulatively (~82 GW of it since January 2025), yet only ~2 GW was online by mid-2026 (Cleanview / SemiAnalysis, 2026). The announcements are a portfolio bet on speed; the thin operating base is the reminder that bridge power is a schedule instrument, not a finished plant. → on-site generation engineering in Chapter 3.5.

Whatever the physical supply structure, most large operators layer a power purchase agreement on top to fix cost and back a clean-energy claim, and the PPA's structure decides whether you have hedged anything at all. The first fork is physical versus virtual. A physical PPA transfers energy to you at an agreed delivery point, which is often the generator's bus rather than your node. Title to the renewable attributes is a contract term, and so is the price. Specify delivery point, shape, shortfall obligations, losses, congestion and REC ownership. The contract fixes only the volumes and prices it explicitly covers; transmission, scheduling, balancing and replacement energy sit outside them unless you buy them too.

A virtual PPA (VPPA) is a financial contract-for-differences. You do not take the power: you settle the difference between a fixed strike and the floating market price at the project's node, and you buy your actual power from the grid separately. A VPPA lets a buyer support a renewable project anywhere, even on a different grid from your data center, and claim the renewable energy certificates, which is why it dominates corporate clean-energy procurement. But it concentrates basis risk. The hedge settles at the project's node, you pay for power at your node, and those two prices diverge. A physical PPA delivered away from your node leaves you the same exposure, because the protection comes from the delivery point, not from the word physical.

The second fork is fixed versus indexed. A fixed-price PPA locks $/MWh for the term. That gives maximum cost certainty, but you forgo any benefit if market prices fall and over-pay against a cheaper future. An indexed (or collared, or hub-settled) PPA floats with a market reference inside a band. It gives less certainty, but it tracks the actual cost of power and avoids the worst over-pay outcomes. For a power-bound AI build the instinct is to fix everything, because Chapter 3.3 showed power is the dominant opex line. But fixing the price at the wrong node, on a project whose generation profile does not match your 24/7 load, can create a hedge that loses money exactly when you need it. A PPA relocates risk rather than reducing it; a badly shaped one relocates it toward you.

Price physical delivery and financial settlement over the same metered intervals. The worked case below isolates the uncovered-hour price at which the better contract flips.

Physical and financial structures can leave the same bill exposure. Retail cash is load × retail price, summed over the intervals. Define the buyer-paid financial settlement as output × (strike − project price) and add it to retail cash. Generator cash is project output × project price plus that settlement. Physical cash instead adds output × strike, output × (retail − project price), and uncovered load × retail price; do not add a VPPA settlement to it. Divide either buyer total by delivered load to compare the energy rate.

Choose the firm offer in the base energy-cost case and either PPA in the lower uncovered-price case, pricing attributes separately. The crossover is where the common covered-interval cash plus uncovered MWh × uncovered price equals the firm-offer cash. A solar settlement debit alone neither proves a bad hedge nor prices the dark hour. The hedge reduces to LP + L(K − P) = LK only when output matches load interval by interval and the settlement and retail prices are identical. The assumed REC transfer covers project generation, not both load intervals.

EPA’s financial-PPA guidance separates retail from settlement, and its physical-PPA guidance covers title and contractual attribute rights. Chapter 15.3 owns hourly matching, and Chapter 3.3 supplies the remaining bill components. Keep basis, shape and collateral in the finance downside.

Write the attribute requirements into procurement at the same time as price: ownership, retirement, eligible geography, vintage, additionality, and the matching interval the customer promises. Additionality means the contract enables new generation rather than buying attributes from an existing project. An annual certificate purchase and an hourly clean-power commitment are different products. Chapter 15.3 defines the accounting and matching method; use its outputs here to price the firming and uncovered hours that the promised matching interval requires. Plan on the lag. Google reports that US interconnection delays often put four to five years between signing a clean PPA and delivery, and its hourly CFE slipped from 66% to 65% in 2025 as its electricity demand grew 37%. A matching promise for a campus energizing within three years should assume existing or already-built supply.

An operator with merchant exposure (anyone drawing grid power without a fully firming physical hedge) is running a power-trading book whether it knows it or not. The risks are specific and quantifiable, and four matter most.

Nodal / locational basis risk. Wholesale dispatch produces locational marginal prices (LMPs) that differ from node to node because of congestion. Suppose your VPPA settles at a renewable project's resource node in West Texas while your Dallas load is priced at a Load Zone SPP or a retail supply product. Congestion can drive the two apart, and that divergence is the basis: an uncovered exposure even though you are nominally hedged. In ERCOT, where renewable generation is concentrated far from load, basis can dwarf the energy price itself. Congestion-revenue rights (CRRs / FTRs) can hedge the congestion component between eligible settlement points, but not retail tariff components or shape. They also reach only about three years forward (ERCOT's long-term auctions, PJM's long-term FTRs), so they leave most of the basis on a 15–20-year VPPA open. Model the actual resource-node-to-load-zone or hub path that matches the contract.

VPPA shape / settlement risk. Beyond basis, a VPPA's generation profile rarely matches your load profile. A solar VPPA generates at midday and settles against you (the negative-settlement trap above), while a 24/7 load consumes flat. That mismatch, the shape, is an uncovered position even when the basis is hedged. The mitigants are storage pairing, hybrid wind-plus-solar portfolios and explicit shape modeling.

Heat-rate / spark-spread risk. For an operator running on-site gas (BTM or bridge), the cost of power depends on the gas price and the plant's heat rate. A heat-rate hedge (or tolling structure) fixes the conversion of fuel to power, not the price of the fuel. At 7 MMBtu/MWh, gas moving from $3 to $6/MMBtu still takes fuel cost from $21 to $42/MWh, and a toller commonly procures its own gas. Insulating the budget from a gas-price spike takes a separate commodity, basis and transport hedge, which matters because the BTM gas wave has tied a large slice of new AI capacity directly to natural-gas markets. Industrial gas-fired LCOE is project-specific. Model heat rate, gas price, capacity factor, capex and financing, O&M, emissions and permits, and interconnection and standby costs before comparing it with ERCOT grid prices. Without a deliberate hedge, the on-site operator stays exposed to power-price volatility.

A PPA without a basis hedge, a shape hedge, and (for gas) a heat-rate hedge is a partial hedge that can lose money in the tail. Map every leg of the supply portfolio to the risk it covers and the risk it leaves open, and price the residual. → the ERCOT-merchant downside is stress-tested in Chapter 1.8; lender downside in Chapter 2.5.

The carbon target and the firmness requirement meet in the search for firm clean supply: power that is both carbon-free and available every hour. There are three paths, on very different timelines. Existing-plant nuclear restarts and uprates are the largest block of new firm clean supply deliverable this decade, though not the only one. Existing hydro and geothermal offtake, and enhanced-geothermal additions such as Google's Fervo projects and its approved 115 MW Nevada agreement, compete for the same role wherever the resource exists. The restart template is Three Mile Island / Crane (835 MW, a 20-year Microsoft PPA, ~$1.6B, backed by a $1B DOE loan). Duane Arnold (615 MW for Google, with a $1.9B DOE loan closed in September 2026) is the second, targeting 2029. The other existing-plant route is long-term offtake from operating reactors, often with uprates and license extensions: Amazon–Talen's 1.92 GW front-of-meter Susquehanna PPA through 2042, and Meta–Vistra's ~2.6 GW, which includes 433 MW of uprates. In all, hyperscalers have announced up to ~13 GW of nuclear, about half from existing plants and half contingent on new reactors (Carnegie Endowment, June 2026). These existing-plant deals work because the plant already exists and is already interconnected; the speed comes from reusing a finished asset, not from anything fast about nuclear.

Small modular reactors (SMRs) such as Kairos (~500 MW for Google), X-energy and Oklo are the 2030s answer, not a near-term speed fix. First-of-a-kind capex runs ~$6,000–13,000/kW overnight (DOE Liftoff, Sept 2024; EPRI models SMR FOAK at $8,460/kWe), and realistic delivery sits in 2032–2035, gated by NRC licensing. Renewables-plus-storage firming is the third path: over-build wind and solar, add batteries (and increasingly long-duration storage), and engineer toward a 24/7-CFE profile without nuclear. It gives the cheapest clean MWh but is the hardest to make firm in the dark, windless hours, which is where the over-build and storage cost lives.

The choice is a timeline bet. If you underwrite nuclear or SMR firm supply as bankable near-term capacity, you may strand a campus waiting for a 2032 reactor. If you treat it as optionality and firm the gap with gas or storage, you carry the bridge cost and the emissions exposure in the meantime. As of 2026, existing-plant restarts are real and bankable where you can secure one; SMRs are optionality, not a schedule input; and renewables-plus-storage is real, but its firmness is only as good as its over-build ratio and its storage duration.

Ask for the deliverability date, not the restart date. Constellation targets a late-2027 restart for Crane, but the PJM transmission upgrades needed for its full output run to December 2030. It covered the gap by moving 760 MW of capacity interconnection rights from its Eddystone plant (FERC waiver, June 2026, over the market monitor's objection). An offtake that includes capacity needs the generator's CIRs in hand for every contract year it relies on. → on-site and nuclear engineering depth in Chapter 3.5.

Co-location means siting the load directly with a generator and drawing power on the generator side of the interconnection point. It promises the most (queue-bypass speed, plus firm clean power if the generator is nuclear), and it lived in the deepest regulatory grey zone until late 2025. The open questions were real. If a data center draws power that never touches the public grid, should it pay for the transmission system it occasionally leans on? Who carries the cost when the co-located load needs grid backup during a generator outage? And does pulling a large nuclear plant off the grid to serve one customer harm the ratepayers who relied on that capacity for reliability and price?

The FERC PJM order of December 18, 2025 answered the first cut of these. It found PJM's existing tariff "unjust and unreasonable" for lacking terms for generators serving co-located load. FERC directed PJM to create three transmission services for co-located load: interim non-firm service while the upgrades for full network service are built, Firm Contract Demand (reserved for at least one year) and Non-Firm Contract Demand (one hour to one month). A co-located load can now contract for a defined slice of grid capacity rather than free-ride or be excluded. PJM filed its compliance tariff in February 2026. On 18 June 2026 FERC granted rehearing in part, accepted that filing only in part, set the services' rates, terms and conditions, and required further tariff language within 60 days. Existing behind-the-meter generation keeps its netting treatment through a three-year transition.

The December order also reformed behind-the-meter generation cost-causation to shield ratepayers. It signaled that Capacity Interconnection Rights (CIRs), the generator's right to inject capacity into the grid, must be adjusted when that capacity is redirected to a co-located load. The orders are PJM-specific. On the same June day FERC put the same questions to the other five FERC-jurisdictional RTOs and ISOs (MISO, SPP, CAISO, ISO-NE, NYISO) through Section 206 show-cause orders, with co-location terms one of five reform areas. ERCOT sits outside FERC's reach: in Texas, SB6 (PURA §39.169) requires ERCOT and PUCT approval before an existing grid-facing generator nets against a new large load.

For a co-location strategy, the regulatory path is no longer a grey zone, but it is now a priced path. You contract for the grid service you use, you face cost-causation charges that earlier deals avoided, and the CIR treatment can change the generator economics that underpinned your offtake. The cautionary case is Talen–Amazon at Susquehanna. When FERC rejected the expanded co-location interconnection agreement in November 2024, the parties re-cut the deal in June 2025 as the front-of-meter PPA described above, with Talen as retail supplier and PPL delivering over the grid, which put transmission charges back in the bill. Draft a co-location offtake with that front-of-meter fallback already priced. → grid-impact process mechanics in Chapter 3.2; grid-services revenue in Chapter 15.8.

The deepest tension in energy supply is a duration mismatch. The firm clean deals that anchor a campus run 15–20 years (Microsoft–Constellation at 20 years, Amazon–Talen at 17), and a behind-the-meter plant or a co-location interconnection is a 20–30-year physical asset. But the GPU fleet those electrons power is tested against a contested 2–3-year economic-life bear case, the sensitivity that Chapter 1.8 turns on. You are committing to power for a fifth of a century to serve a chip whose economic life may be that short (published useful lives run 4–6 years) and a workload that might move in five. A take-or-pay power contract or a co-location agreement does not care whether the GPUs behind it are still earning revenue. If the workload migrates, demand softens, or the residual market for the hardware collapses, the power commitment remains: a fixed liability against a vanished revenue stream.

The mitigants are all forms of optionality. Match the firm tenor to the durable part of the load, not the speculative part. Contract long-dated power only for capacity you are confident will run for the term, and meet uncertain incremental demand with shorter, more flexible supply (grid, spot or curtailable). Grid service is not automatically the flexible leg, though. Large-load tariffs now carry take-or-pay terms of their own. Dominion's GS-5 class (from 2027) requires 14-year contracts that bill at least 85% of transmission-and-distribution demand and 60% of generation demand. AEP Ohio's data-center tariff sets the terms in Chapter 2.4. Compare exit fees, ramp schedules and collateral before treating grid as the shorter tenor. Keep the supply portfolio re-mixable. A co-location anchor plus flexible grid service plus a storage layer can shed or shift cost as the workload changes; a single 20-year fixed-price BTM commitment cannot. And price the power against the economic GPU life, not the contract life. If the energy strategy only pencils when amortized over 20 years of GPU operation, it does not pencil, because the GPUs will not be there. Size the irreversible firm commitment to the workload you are certain of and let reversible, flexible supply absorb the rest, the same reversibility sort Chapter 1.8 applies to the fleet.

The recurring energy-supply mistakes all come from treating a risk-relocation instrument as a cost-reducer, or from matching the wrong durations:

• The unhedged-basis VPPA. Signing a virtual PPA at a remote renewable node to "fix" power cost, with no plan to hedge congestion between the project node and the load node. CRRs/FTRs cover only a few years at a time, not the whole term. The VPPA settles where the project is and you pay where the load is; the gap is an uncovered position that blows out when congestion is worst. It is a partial hedge sold internally as a full one.
• The single-technology solar VPPA on a 24/7 load. Buying a solar-only VPPA to power a flat round-the-clock load, then discovering it settles against you through the high-solar hours (cannibalization) and supplies nothing in the dark hours. The negative settlement is not itself the loss, since that debit is how a contract for differences pays the fixed price. The loss is the residual: retail cost plus settlement, hour by hour, at your node rather than the project's. Model that sum before assuming a wind-weighted portfolio has fixed either shape or basis.
• The permanent bridge. Building on-site gas as a bridge to grid power, then watching turbine lead times and queue delays stretch until the "bridge" is the permanent plant, now carrying fuel risk, emissions liability and stranding risk it was never sized to hold.
• The tenor mismatch. Signing a 17–20-year firm power contract sized to a peak load that depends on a workload whose earning life may be as short as the contested 2–3-year bear case, so a workload migration leaves a take-or-pay liability against a campus that no longer earns. → Chapter 1.8.
• The grey-zone co-location pro-forma. Underwriting a co-location deal on pre-order economics, where the load free-rode on transmission and CIR value, a model that FERC's PJM orders of December 2025 and June 2026 now let a regulator unwind.