On-Site & Bring-Your-Own-Power Generation (Energy-Supply Strategy)

For two decades, on-site generation in a data center meant a yard full of diesel gensets that ran a few hours a year for backup on a NFPA test schedule. By 2026, on-site and bring-your-own-power (BYOP) generation has flipped from a backup-only afterthought to the primary energy-supply strategy for new gigawatt-scale AI campuses — not because operators want to run power plants, but because the grid cannot energize them on the schedule their capital plan assumes. When the binding constraint is megawatts-and-when rather than chips-or-capital (Chapter 3.2), the rational move is to stop waiting in the queue and generate your own electrons. This chapter is the energy-supply decision made under that constraint.

It is deliberately a strategy chapter, not an engineering one. The electrical integration — sizing, paralleling, step-loading into synchronized GPU pulse loads, protection coordination, grid-forming inverters — is its own discipline, treated in Chapter 4.8; the gas-process and fuel-conditioning engineering lives in Chapter 4.9. Here we work the forks a strategist and a lead engineer must agree on before any prime mover is ordered: bridge vs permanent, speed vs efficiency, islanded vs grid-parallel, firm vs interruptible fuel, and which emissions envelope governs the whole thing.

The economics start with a single number: a gigawatt of AI capacity is worth roughly $10–12B/year in revenue (SemiAnalysis, 2025). Run that backward and the value of speed is direct — getting 200 MW online six months early is worth on the order of $1–1.2B in pulled-forward revenue against a depreciation clock that is already running on the GPUs inside. Against a grid queue where PJM application-to-COD now exceeds eight years (versus under two in 2008) and ERCOT screened on the order of 198 GW of large-load applications in Q1 2026 against roughly 1 GW of trailing-twelve-month approvals, the decision to self-generate is not ideology. It is the only way to convert a powered shell into a revenue-earning asset on the timeline the silicon demands.

That produces the bridge-power thesis and a three-phase deployment sequence that has become the industry default. Phase 1 (now): simple-cycle gas — reciprocating engines and aeroderivative turbines — plus BESS, deployed in 12–24 months to get first power while the grid request matures. Phase 2 (2026–2028): add efficiency where the plant will live for years — steam tails / combined cycle, or solid-oxide fuel cells — and firm up fuel supply. Phase 3 (2027–2035): existing-nuclear restarts and uprates arrive this decade; SMRs and 24/7 carbon-free renewable-firm hybrids arrive in the 2030s; on-site gas demotes to backup as the grid and clean-firm supply finally catch up. The trap is in the seam between the phases: a bridge plant that cannot be cleanly demoted — because the grid connection never arrived, or because its air permit assumed temporary operation — becomes a permanent plant you under-engineered, or a stranded asset you cannot recover. Whether you are building a bridge or a permanent plant is the first fork, and it governs every choice below it.

Once you know whether you are buying speed or efficiency, the prime-mover choice falls out of three variables: lead time (how fast can you get the iron), efficiency / heat rate (what you burn per MWh), and permitting posture (what the plant emits and how hard that is to permit). The four combustion options plus fuel cells and nuclear span the entire tradeoff space. The table below is the decision, not a catalog — read it as 'what do I give up to get what I need.'

Reciprocating engines (RICE) — Wärtsilä, INNIO Jenbacher, Caterpillar, Bergen/Everllence — are the speed-and-flexibility play. They deliver the highest simple-cycle efficiency of any combustion technology (45–48%), start to full load in about two minutes, ramp faster than 100%/minute, and turn down to ~10% load — which makes them the natural match for the spiky, synchronized pulse loads an AI cluster throws at its power plant. The cost is modularity-as-burden: you need roughly 200+ units to make 500 MW, which is a lot of paralleling, switchgear, and O&M headcount. Aeroderivative turbines (GE LM2500 ~34 MW, LM6000 ~52–57 MW; refurbished ProEnergy PE6000 from repurposed jet cores) are denser per unit and start in 5–10 minutes, but run lower simple-cycle efficiency (37–42%), derate hard in heat, and bottleneck on turbine-blade casting supply. Heavy-frame turbines and combined cycle push past 60% efficiency in CCGT configuration but carry 3–5 year lead times (sold out into 2028–29) and 24–36 month installs — the efficiency endgame, not a bridge.

Combined cycle, CHP, and thermodynamic optimization are an energy-strategy choice, not a default. Bolting a steam tail onto simple-cycle exhaust lifts a 42% plant toward 60%+ — but it adds a steam turbine, a heat-recovery steam generator, a condenser, and a water demand that collides directly with the water-availability siting gate (Chapter 3.7). For a true bridge plant that demotes to backup in three years, the steam tail is capital you mothball. For a permanent plant, it is the difference between competitive and uncompetitive delivered energy. The decision hinges entirely on the bridge-vs-permanent fork above — which is why that fork has to be settled first. There is a tempting middle path — combined heat and power, where reject heat warms an adjacent load — but in most AI siting there is no co-located thermal host, so CHP economics rarely close unless heat reuse or district heating is itself a siting feature (Chapter 15.4).

Nuclear is the one option whose strategic role is most often mis-stated. It splits into two tracks with very different timelines. Existing-plant restarts and uprates deliver clean firm power this decade: the Three Mile Island / Crane restart (835 MW for Microsoft, targeting ~2027–28), Palisades, Clinton — these are real megawatts on a near-term clock because the asset already exists and the licensing path is a restart, not a new build. SMRs (Kairos, X-energy, Oklo, NuScale) are a 2032–2035 scale story with first-of-a-kind capex of $6,400–12,700/kW and licensing risk that has not been retired. The strategic error is treating either track as a speed fix for a 2026 power need. Restarts are a scarce, largely one-time inventory of clean-firm capacity that a handful of hyperscalers have already locked up; SMRs are a hedge you can underwrite as optionality but should not put on the critical path. Deloitte's own framing has new nuclear at roughly 10% of data-center demand growth by 2035 — meaningful, late, and clean, but not the answer to 'I need 500 MW in 2027.' For the firm-clean-supply procurement structures (17–20 year PPAs, take-or-pay, FERC co-location treatment), see Chapter 3.4.

The clean-but-firm ambition runs into a hard physical fact: to make an off-grid solar/wind plant deliver firm 24/7 power, you over-build the generation by roughly 7x (a contested, single-source figure sensitive to site and storage assumptions) and back it with very large storage — and even then the cheapest sites land near $113/MWh firmed (2025), trending toward ~$77/MWh by 2030. That overbuild ratio is why pure renewables-off-grid rarely anchors a gigawatt AI campus on its own today; the land, the capex, and the storage all balloon. Where renewables-anchored hybrids do work is as the clean fraction of a firmed portfolio — solar/wind for the energy, gas or fuel cells for the firmness, BESS for the seconds-to-minutes ride-through and the inertia an islanded low-rotating-mass plant otherwise lacks. This is the 24/7 carbon-free energy (CFE) frontier: not 100% renewable instantaneously, but hour-matched clean supply firmed by dispatchable capacity. The decision here is how much clean fraction you are willing to pay the overbuild premium for, against a corporate CFE commitment and an increasingly carbon-aware permitting and community environment (Chapter 3.11 on social license; Chapter 15.4 on stewardship).

The most under-appreciated failure mode in BYOP is to solve the electrical critical path and forget the molecular one. A gas plant is only as firm as its fuel, and firm pipeline transport is scarce. The lateral and midstream buildout to bring sufficient gas to a gigawatt-class site can take as long as — or longer than — the turbines themselves, and INGAA projects the US needs roughly +39% pipeline capacity by 2052 to serve the load. A gas-supply interruption is, functionally, a generation outage — which means fuel firmness is a reliability-engineering problem, not a procurement footnote. This is why the lead-time register in Chapter 2.3 must carry the pipeline lateral as a long-lead item alongside the prime movers and transformers.

The decision is firm vs interruptible transport, and how you synthesize firmness when firm transport is unavailable. Firm transport is contractually guaranteed but expensive and supply-constrained; interruptible is cheaper but can be curtailed precisely when you need it most — during a correlated cold snap, when residential heating load has priority. The practitioner answer is 'synthetic firm': layer multiple pipelines, interruptible transport, on-site LNG/CNG storage, and dual-fuel switching so no single curtailment takes the plant down. LNG/CNG is also the bridge fuel that lets a site run before the pipeline lateral is complete — an LNG truck carries ~826 Dth versus ~450–550 Dth for CNG, and LNG extends the economic trucking radius to roughly 300 miles. The sizing question — how many days of on-site storage hedge a cold-snap curtailment — is a fuel-process engineering problem detailed in Chapter 4.9; the strategic point here is that supply interruption is a firmness risk co-equal with the electrical interconnection, and a BYOP plan that treats fuel as a given has an unhedged single point of failure.

The islanded-vs-grid-parallel fork is as much a regulatory decision as a technical one, and the answer is market-dependent. Fully islanded (never touches the grid) is the cleanest regulatory path — you avoid interconnection queues and most FERC/ISO jurisdiction entirely — at the cost of the full overbuild obligation above and zero grid backstop. Grid-parallel / co-located uses the grid as N+1 and can sell or buy at the boundary, but it pulls you into the co-location rulemaking that crystallized in 2026. FERC's December 18, 2025 order directs PJM to create three co-location transmission services (Firm, Non-Firm, and Interim Non-Firm Contract Demand) with a transition running to December 2028 — a framework likely to template across MISO, SPP, and ERCOT. The live controversy is cost-causation: who pays for the grid backstop a co-located load relies on, and how to prevent the cost from shifting onto other ratepayers (a concern estimated around $140M/yr in some markets). The market you site in changes the answer entirely: ERCOT's lighter-touch, energy-only structure differs sharply from PJM's capacity-market co-location order and from a vertically-integrated regulated utility where the PUC governs the tariff. The regulatory-treatment detail and PPA structures sit in Chapter 3.4; the point here is that where you island determines whether islanding is even the cheaper path.

Here is the recurring irony of bridge power: operators turn to on-site gas to escape a multi-year grid wait, and then run headlong into a multi-year air-permit wait. A combustion plant of any size is a stationary source; cross the thresholds and it becomes a major source with PSD/Title V obligations, GHG reporting, and — in a nonattainment area — major-source-style BACT/LAER analysis, dispersion modeling, and community health-risk assessment even for nominally 'emergency' fleets. Air permits often exceed a year even in permitting-friendly Texas. The schedule advantage you bought with fast turbines can be eaten whole by the permit, which is why the air permit belongs on the critical-path register (Chapter 2.3) and why the full federal/state/local permitting matrix is treated in Chapter 3.9.

Two 2026 developments sharpen the fork. First, EPA's combustion-turbine NSPS tightened the 'temporary turbine' subcategory — historically used to run bridge turbines under a lighter permit on the theory they would be removed (the subcategory contemplates units under ~850 MMBtu/h operating ≤24 months at a site). A bridge plant permitted as temporary that the grid connection never relieves faces retroactive permit-cliff and citizen-suit risk: the very assumption that made it fast to permit becomes a liability when 'temporary' turns permanent. Second, this is exactly why fuel cells' near-zero criteria-pollutant profile is strategically valuable — they sidestep the combustion-permit envelope entirely. The emissions decision is an input to the prime-mover choice and to the bridge-vs-permanent fork at the very top of the tree, not a compliance afterthought bolted on at the end. Water and discharge permitting for any steam-cycle or evaporative-cooling component compounds it (Chapter 3.7).

The energy-supply strategy is the product of four nested decisions, taken in order. (1) Bridge or permanent? — sets your efficiency tier and permit posture. (2) Speed or cheap energy? — sets your prime mover (engine/aero for speed; CCGT/nuclear for cost). (3) Islanded or grid-parallel? — sets your overbuild obligation and your FERC/ISO exposure. (4) Firm or synthetic-firm fuel? — sets your second critical path and your cold-snap resilience. Each answer constrains the next, and each has a downstream cost that lands years later: the mothballed steam tail, the stranded high-heat-rate bridge, the cost-causation surprise, the curtailed pipeline. The operators getting this right in 2026 are not the ones with the cleverest prime mover — they are the ones who settled bridge-vs-permanent honestly at the top, and let every downstream choice inherit from it.

None of this displaces the grid as the eventual destination. BYOP is the response to a queue that will, eventually, clear; the forward roadmap demotes on-site gas to backup as grid upgrades, queue reform, existing-nuclear restarts, and clean-firm hybrids arrive across 2027–2035. The strategic posture is to use BYOP to win the speed race without over-committing to a plant the grid will later make redundant. The grid-supply alternatives that BYOP is racing against are detailed in Chapter 3.3 and Chapter 3.4.