Stranded Energy to Sustainable AI: Inside Crusoe’s Abilene, TX Data Center Campus
This is Part 2 of a 3-part blog series around the looming climate challenge: the massive surge in data center energy consumption driven by AI. See Part 1 outlining the opportunity in stranded energy and efficient compute here. Written by the team at G2 Venture Partners.
AI data centers are expanding fast, with over 100 GW of new capacity planned in the U.S. from 2024–2035 – 5x today’s 20 GW¹. U.S. utilities estimate data center power demand could reach 900 TWh by 2035, up from ~185 TWh in 2023².
To meet this demand, natural gas has become the default power source due to its reliability, speed, and cost-effectiveness. This reliance is driving new gas plant development, locking in fossil fuel dependence for decades.
But it doesn’t have to be this way. Crusoe, a G2 Venture Partners portfolio company, is taking a different approach — siting data centers in regions with the most abundant renewable energy resources. This enables both (i) the use of stranded or curtailed energy and (ii) the buildout of more renewables in these geographies. For example, Crusoe’s 1 GW+ Abilene data center development in West Texas runs on excess wind power that would otherwise be wasted. By colocating compute in regions best-suited for renewable development, Crusoe is proving that AI’s rapid expansion doesn’t have to come at the cost of sustainability.
Status Quo: Natural Gas as the Backbone of New Data Center Power
Natural gas has become the default choice for powering the AI data center boom — not because it’s the cleanest option (it isn’t) but because for most utilities and data center operators, it’s the easiest, fastest, and most economical way to get firm power at scale. Key factors driving this preference include:
- Reliability: Data center workloads require 99.9%+ uptime, and natural gas provides stable, firm power without the intermittency of wind or solar.
- Speed: In the “AI arms race”, companies need compute capacity, yesterday. Modular natural gas turbines can be procured relatively quickly, though they now face a significant backlog, while new gas plants take 2–3 years to build and major transmission or nuclear projects take much longer.
- Cost-Effectiveness: While wind and solar are cheaper per MWh, they require higher-cost storage or backup power for 24/7 reliability. Many operators still opt for gas due to the complexity of scaling a renewables solution.
- Ease: The U.S. has an extensive natural gas network, making scaling gas infrastructure simpler than building entirely new renewables or nuclear capacity.
Experts posit that, in the medium term, 75–100% of incremental data center power needs will be supplied by natural gas. In practice, this means most new hyperscale facilities are planning to rely on gas-fired electricity, whether via grid power or on-site generation. This trend is already evident, as power producers and oil & gas majors have provisioned tens of GWs of new gas-fired power plants primarily to meet data center demand⁴⁵.
This deepening dependence on natural gas raises concerns about carbon emissions and sustainability. Even as hyperscalers set long-term carbon-free energy goals, in practice many are turning to gas in the interim because truly clean firm power isn’t widely available. The result is a near-term surge in carbon-intensive generation — a dynamic prompting interest in lower-emission, and lower-cost, alternatives such as “stranded” energy.
The Opportunity: Leveraging Clean, Stranded, and “Negative”-Priced Electricity
As renewables expand, some U.S. regions generate more electricity than the grid can absorb, leading to curtailment — where wind and solar power is wasted rather than delivered to consumers. This is most pronounced in high-renewable energy regions like Texas (ERCOT), the Midwest (SPP, MISO), and California (CAISO).
- In Part 1, we highlighted how ERCOT curtailed 5% of wind and 9% of utility-scale solar in 2022 — approximately 55 GW in total. By 2035, curtailments could double to 110 GW⁶.
- In SPP, wind curtailment rose from 2.7% to 6.4% from 2020 to 2021⁷.
- CAISO curtailed 2.4 TWh of renewable energy in 2022 — a 63% increase from the previous year⁸.
Why do power producers curtail? For two primary reasons:
- Geographic or congestion curtailment: Limited transmission capacity to distribute power supply to demand.
- Temporal curtailment: Periods of low demand coinciding with peak renewable output.
By 2035, 36% of curtailment in ERCOT is expected to be directly tied to transmission congestion⁹. This excess energy often drives wholesale electricity prices to zero or even negative — in SPP, negative prices occurred in over 25% of hours in wind-heavy areas in 2020¹⁰, with congestion causing ~60% of these events in 2023¹¹. In ERCOT, system-wide negative prices rose from 110 hours in 2022 to 139 hours in 2023¹².
This creates a major opportunity: data centers colocated with stranded renewable energy can access ultra-low-cost power while cutting emissions. Instead of relying on new gas plants, AI infrastructure should be built where energy is abundant, cheap, and clean. As curtailment grows, utilizing this wasted power will become even more compelling. And in the long run, siting data centers in high-potential renewable energy regions will drive the development of additional low-cost renewable capacity. Conversely, if data centers are built in areas more suitable for natural gas expansion, fossil fuels will remain the dominant power source for decades to come.
The Solution: Crusoe
Crusoe’s Abilene, TX project demonstrates this approach. Instead of defaulting to solely natural gas, it taps into excess renewable energy, aligning AI’s expansion with a low-carbon strategy. When completed, the Abilene development will provide 1 GW+ of AI computing capacity, making it one of the largest AI-dedicated data center campuses in the U.S.
By colocating its data center in a stranded wind energy hub, Crusoe captures and uses excess power for AI workloads, and provides the demand to enable new Gigawatt-scale wind farms to be economically built. This approach enables new renewable power plants, improves renewables utilization, stabilizes the grid, and delivers reliable, low-cost compute while reducing emissions.
Importantly, the rapid growth that utilities and market participants project may not be permanent — some anticipate that after 2030, the shift to inference and inevitable consolidation will slow or even reduce new data center additions. Investing in sustainable solutions now prevents overbuilding costly gas plants that could become obsolete as demand stabilizes post-2030.
At G2, we aim to do more with less. The most sustainable path forward is to fully utilize the energy we already generate rather than defaulting to new carbon-intensive infrastructure. By colocating AI data centers with stranded renewable energy, as Crusoe is doing in Abilene, we can scale computing while avoiding decades of emissions and driving the development of additional low-cost renewable capacity. This approach reduces waste, lowers costs, and aligns AI’s growth with a cleaner energy future.
Sources:
¹ https://www.rystadenergy.com/insights/data-centers-reshape-us-power-sector
³ TD Cowen, Data Centers, Generative AI & Power Constraints: The Path Forward, May 2024.
⁵ https://www.rystadenergy.com/insights/data-centers-reshape-us-power-sector
⁶ https://www.eia.gov/todayinenergy/detail.php?id=57100
⁷ https://www.energy.gov/sites/default/files/2022-08/land_based_wind_market_report_2202.pdf
⁸https://www.eia.gov/todayinenergy/detail.php?id=60822
⁹https://www.eia.gov/todayinenergy/detail.php?id=57100
¹⁰ https://emp.lbl.gov/news/berkeley-lab-study-investigates-how
¹¹https://www.eia.gov/electricity/wholesalemarkets/index.php
