Terrestrial Power-Asset Capital Cost
What is the power-generation capital cost per kW_IT for behind-the-meter (BTM) generation at terrestrial AI data centers?
What is the power-generation capital cost per kW_IT for behind-the-meter generation at terrestrial AI data centers?
Answer
BTM power-generation capex per kW_IT ranges from $150-$325/kW_IT in 2026 to $500-$900/kW_IT by 2040, driven by rising BTM penetration and evolving generation technology mix. This is a modest cost component: at the central 2030 estimate of $500/kW_IT, amortized using CRF(7%, 20 years) = 0.094, it adds ~$47/kW_IT/year — roughly 0.4% of total TCO.
The optimistic scenario has the highest power-asset capex because it assumes the most aggressive BTM deployment (70% by 2040). The conservative scenario has the lowest because it assumes operators remain mostly grid-connected (28% BTM by 2040). This counterintuitive pattern — where low-energy-cost scenarios require higher capex — is precisely the completeness issue this page addresses.
Central estimates ($/kW_IT, one-time capex, 2025 USD):
- 2026: $200 (10% BTM, mostly gas)
- 2030: $500 (28% BTM, gas + early solar+storage)
- 2035: $650 (40% BTM, diversifying mix)
- 2040: $750 (45% BTM, mature mix including SMR)
Analysis
Generation technology capex ($/kW_gen, 2025 USD)
| Technology | Low | Central | High | Notes |
|---|---|---|---|---|
| Simple-cycle gas (BTM-oriented) | $800 | $1,200 | $1,800 | Boom at $1,033; GridLab filings higher |
| Solar + 4hr battery (co-located) | $1,500 | $2,175 | $2,800 | EIA benchmark; declining per BNEF |
| SMR (NOAK, 2035-2040) | $4,000 | $6,500 | $10,000 | Highly uncertain |
Gas turbine costs have escalated sharply: real project costs from IRP/CPCN filings are ~2x EIA baseline assumptions gridlab-gas-turbine-costs-2025.1, driven by data center demand competing for turbine supply. BTM gas typically uses simple-cycle or aeroderivative turbines at $800-$1,800/kW_gen.
BTM penetration rate estimates
| Year | Optimistic | Central | Conservative |
|---|---|---|---|
| 2026 | 15% | 10% | 8% |
| 2030 | 40% | 28% | 18% |
| 2035 | 60% | 40% | 25% |
| 2040 | 70% | 45% | 28% |
Optimistic follows Patel's "half by end of decade" and continued growth. Central tracks McKinsey 25-33% for 2030 with modest extension. Conservative reflects operator preference to return to grid-tied latitude-btm-traction.3.
BTM generation mix (% of BTM capacity, central scenario)
| Technology | 2026 | 2030 | 2035 | 2040 |
|---|---|---|---|---|
| Gas | 90% | 75% | 50% | 35% |
| Solar + storage | 10% | 22% | 38% | 40% |
| Nuclear | 0% | 3% | 12% | 25% |
Converting to $/kW_IT
power_asset_capex = BTM_share × PUE × weighted_avg_capex_per_kW_gen
The PUE multiplier accounts for BTM generation covering total facility power, not just IT load. However, not all BTM capex appears as operator-owned capital — some is structured as third-party PPA/lease, shifting capex off-balance-sheet into energy opex. The values reflect the net effective capital exposure.
Relationship to energy cost page
The terrestrial-energy-cost page's blended rates include BTM LCOE, which embeds amortized generation capex. This page separates out that capex component for transparency. The corresponding variable-only energy costs (fuel + O&M, excluding BTM capex amortization) are lower than the blended rates:
| Year | Central blended | Central variable-only | Difference |
|---|---|---|---|
| 2026 | $0.075/kWh | $0.072/kWh | $0.003 |
| 2030 | $0.080/kWh | $0.073/kWh | $0.007 |
| 2035 | $0.075/kWh | $0.066/kWh | $0.009 |
| 2040 | $0.070/kWh | $0.060/kWh | $0.010 |
The sum of variable energy cost × 8760h × PUE + power_asset_capex × CRF(7%, 20yr) approximately equals the blended energy cost × 8760h × PUE from the original model, confirming the decomposition is consistent. Note that using simple 1/20 amortization instead of CRF would understate the annual cost by roughly 47% ($25 vs $47/kW_IT/year at $500 capex) — the model correctly uses CRF to reflect the cost of capital.
Evidence
A. Gas Turbine Capital Costs
A1. Boom Superpower: 42 MW gas turbine at $1,033/kW_gen. Targets 39% efficiency; maintains nameplate at 110°F+ without water. — boom-superpower-turbine
A2. GridLab 2025 analyzed IRP/CPCN filings: simple-cycle projects (2025-2029) at $728-$1,965/kW_gen; CCGT projects (2026-2031) at $1,116-$2,256/kW_gen. Recent CCGTs routinely exceed $2,000/kW_gen due to demand-driven turbine pricing power. — gridlab-gas-turbine-costs-2025
A3. EIA/Sargent & Lundy (AEO2025): H-class CCGT at ~$921/kW_gen base; aeroderivative simple-cycle at ~$1,606/kW_gen. Location adjustments 0.98-1.21x. — eia-capital-cost-aeo2025
A4. Lazard estimates CCGT capex at ~$2,000/kW_gen and OCGT at ~$1,300/kW_gen (LCOE+ 2025). — lazard-lcoe-2025
B. Solar + Storage Capital Costs
B1. EIA/Sargent & Lundy: utility-scale solar PV + 4-hour battery (150 MW solar / 50 MW battery) at $2,175/kW_gen net. — eia-capital-cost-aeo2025
B2. BNEF: 4-hour battery LCOS fell 27% to $78/MWh. Solar+storage deployed at $57/MWh in 2025. BNEF forecasts 30% solar LCOE reduction and 25% battery reduction by 2035. — bnef-lcoe-2026
B3. NREL 2025: 4-hour battery at $334/kWh ($1,336/kW power capacity). Projections: $147-$339/kWh by 2035. — nrel-battery-cost-2025
C. BTM Penetration
C1. 56 GW BTM planned for US data centers (~30% of projected capacity). McKinsey: 25-33% of incremental demand via BTM through 2030. — latitude-btm-traction
C2. Major BTM gas: Stargate 7 GW, Joule 1.3 GW islanded, VoltaGrid 2.3 GW for Oracle, FO Permian/Hivolt 5 GW. — latitude-btm-traction
C3. SemiAnalysis: "something like half of the capacity that's being added will be behind the meter" by end of decade. BTM is more expensive than grid but faster due to interconnection queues. — patel-2024-ai-bottlenecks
C4. Schneider Electric CTO notes most major DC companies don't want to own/operate power systems long-term; AWS goal is to "get back to being grid-tied." — latitude-btm-traction
D. SMR Capital Costs
D1. SMR LCOE: FOAK ~$180/MWh; NOAK estimates $60-$80/MWh by late 2030s. Capital costs range $4,000-$10,000+/kW_gen depending on maturity. — introl-smr-timeline