POWER Magazine: 27% of US Data Centers to Run on Onsite Generation
- Onsite Generation Share by 2030: 27% of US data centers (versus roughly 1% previously)
- Self-Generated Capacity by 2030: 35 GW
- US Data Center Demand 2030: 134.4 GW (S&P Global)
- Utility Interconnection Wait: 3 to 5 years
- Wartsila Ohio Project: 282 MW from 15 Wartsila 18V50SG engines
POWER Magazine reported on May 18, 2026, that 27 percent of US data centers are expected to rely entirely on onsite generation by 2030, a significant increase from approximately 1 percent previously. This shift anticipates 35 GW of total data center demand being self-generated rather than sourced from the grid.
The Data
S&P Global Look Forward research projects US data center demand to reach 75.8 GW in 2026 and 134.4 GW by 2030. Data centers already consume about 25 percent of Virginia’s electricity, with PJM forecasting continued growth. Wood Mackenzie reports Texas leads in planned data center capacity using onsite generation, with deployments in Texas, Wyoming, and New Mexico designed to be invisible to the grid. In contrast, onsite generation in PJM states like Pennsylvania and Virginia is more often structured to remain grid-tied behind the meter. Rolls-Royce reported a 50 percent growth in sales of mtu power generation products for the data center segment in 2024. INNIO Group announced a 2.3 GW order with VoltaGrid, its largest ever. Wartsila received a 282 MW order for natural gas reciprocating engines for a project in Ohio. Caterpillar gas engines are being deployed at Monarch Compute Campus in West Virginia (2 GW), a Joule Capital Partners Utah campus (4 GW total), and a Hunt Energy and Caterpillar partnership for up to 1 GW across North America. Bloom Energy is deploying 1.2 GW of fuel cell capacity at Oracle data center sites, with Oracle intending to procure up to 2.8 GW. xAI has deployed up to 35 mobile gas turbines at its Colossus data center in Memphis. Energy Abundance Development Corp announced Data City, Texas, a 5 GW fully off-grid data center hub.
What It Means
The substantial increase in data center reliance on onsite generation stems from utility interconnection wait times of 3 to 5 years, a timeline that clashes with AI data center developers’ need for power availability within 1 to 2 years. Analyst Vlad Galabov noted that “those wanting to build new data centers have discovered that many utilities cannot deliver the power they need for several years.” This disconnect is reshaping the energy landscape, with companies like Monarch Compute Campus and Joule Capital Partners opting for large-scale self-generation to meet their power requirements. The trend also extends to emerging technologies, as seen with Oracle’s substantial fuel cell procurement from Bloom Energy and xAI’s deployment of mobile gas turbines.
The Gap
The POWER Magazine and cited research have not detailed the fuel mix for the 27 percent onsite-only fraction, nor have they distinguished between “behind-the-meter co-located generation” and “fully islanded” deployments. The reported 35 GW self-generated total does not specify firm capacity versus nameplate ratings, nor does it account for backup-only equipment that may remain idle. Wood Mackenzie’s Q4 2025 release indicated a slowdown in newly added US data center capacity, suggesting that interconnection-queue and onsite-permitting bottlenecks are now constraining both grid-tied and off-grid pathways. Furthermore, state regulators in Texas, Virginia, and Wyoming have not yet finalized standardized rules for large-load co-located generation, introducing permitting risk for developers in these active markets.
Critical Perspective
The POWER Magazine report’s assertion that 27% of US data centers will run entirely on onsite generation by 2030 is overly optimistic without specifying the fuel mix or distinguishing between behind-the-meter co-located and fully islanded deployments. According to Watt-Logic’s research, the actual firm capacity from the Wartsila Ohio project, at 282 MW, represents only a fraction of the total self-generated capacity forecasted by POWER Magazine (35 GW). This discrepancy raises questions about how much of the reported 134.4 GW US data center demand in 2030 will truly be met with firm onsite generation. Given that state regulators in Texas, Virginia, and Wyoming have yet to finalize standardized rules for large-load co-located generation, significant permitting risks could delay or even stall these projects. Moreover, the interconnection queue backlog of 1,200 GW suggests that grid-tied capacity will face substantial delays. The question remains: how much of this self-generated capacity is actually firm and not reliant on grid support during peak demand periods?
Why It Matters
The trend of US data centers relying on onsite generation has significant implications for the energy landscape. With 27% of data centers expected to rely entirely on onsite generation by 2030, this shift will impact the demand on the grid, particularly in regions like Texas, Virginia, and Wyoming. The growth of data center demand, forecasted to reach 134.4 GW by 2030, will require innovative solutions to meet power requirements, driving investments in onsite generation technologies.
The increase in onsite generation is largely driven by utility interconnection wait times, which can take 3 to 5 years, clashing with the need for rapid power availability in the data center industry. This has led companies to opt for self-generation, reshaping the energy market and creating new opportunities for technologies like fuel cells and mobile gas turbines. As data centers continue to grow, the demand for firm and reliable onsite generation will become increasingly important.
The implications of this trend extend to broader grid and policy trends, including the role of ERCOT, PJM, and CAISO operators in managing data center demand, the economics of battery energy storage systems (BESS), and the impact of FERC orders on the energy market. As the energy landscape continues to evolve, understanding the trends and challenges in onsite generation will be crucial for US utilities, data center operators, and policymakers.