The Substation Is No Longer the Starting Point
For most of the data center industry’s history, the substation was fixed infrastructure. Operators found a site, identified the utility’s delivery point, and built backward from there. Power capacity arrived from the grid, and facility engineering optimized what happened after the meter. That model assumed utilities could accommodate whatever load was requested within a commercially acceptable timeframe. That assumption is now broken in every major data center market on earth. Northern Virginia’s Dominion Energy interconnection queue stretches to seven years. West London’s distribution network carries load requests that will remain unanswered for at least half a decade. Phoenix, Chicago, and Dallas-Fort Worth face similar dynamics with different timelines. Grid operators across Europe, North America, and the Asia Pacific have confirmed that the velocity of data center load growth cannot be absorbed by transmission infrastructure built for industrial-era demand profiles. The utility substation has not disappeared from the picture.
Oracle’s Project Jupiter: A Design Reversal That Signals an Industry Shift
Oracle’s Project Jupiter in Doña Ana County, New Mexico, arrived in early 2026 already carrying significant attention. The planned AI data center campus spans roughly 1,400 acres and could include four hyperscale buildings as part of a long-term investment program. What changed the narrative was not the scale — it was a design reversal that exposed how quickly the underlying power engineering logic of major AI campuses has shifted. Oracle, in partnership with campus developer BorderPlex Digital Assets, originally planned the facility around a combination of natural gas turbines and diesel generators. The revised announcement made in April 2026 replaced both technologies in their entirety. Bloom Energy’s solid oxide fuel cell technology would instead power the campus as a single on-site microgrid, with installed capacity reaching up to 2.45 gigawatts. Oracle confirmed the system operates without combustion, uses minimal water, and pairs with closed-loop non-evaporative cooling.
Why the Reversal Happened
The shift was not purely technical. Community opposition to the original gas turbine and diesel generator design had generated pressure from environmental groups and local residents in southern New Mexico. Oracle’s revised announcement cited commitment to both innovation and community priorities. That combination of engineering advantage and community relations management reflects a wider reality: data center developers in 2026 face scrutiny from local governments and ratepayer advocates at the same time that grid access is constrained. A behind-the-meter fuel cell microgrid addresses both concerns simultaneously it bypasses congested interconnection queues and produces power without combustion emissions or significant water consumption.
Bloom Energy’s Chief Commercial Officer Aman Joshi described Project Jupiter as potentially becoming one of the largest data center microgrids operating in the United States at the time of completion. He framed it explicitly as “a model that can be replicated across America.” That language matters. Oracle did not characterize the Project Jupiter energy redesign as a one-off solution unique to New Mexico’s regulatory environment. The company positioned it as an architecture applicable across the national data center build-out, which signals that fuel cell microgrid design is moving from pilot-scale experimentation into mainstream gigawatt-scale deployment.
Natural Gas With Carbon Capture
Natural gas-fired generation remains the dominant behind-the-meter technology for new data center projects in 2026. The reason is straightforward: combustion turbines and reciprocating engines can be procured, permitted, and commissioned within 18 months — a window that matches data center construction timelines in ways that nuclear, large-scale renewable, or hydrogen alternatives currently cannot. Woodway Energy’s analysis of the behind-the-meter market frames gas generation explicitly as the bridge while grid infrastructure catches up. CHP microgrids — combined heat and power configurations that capture waste heat from gas turbines to generate additional electricity or provide cooling — achieve efficiency rates significantly above conventional grid power delivery. Intrastate gas pipelines, which operate under state rather than federal regulatory frameworks, can accelerate project delivery further by reducing permitting complexity.
The constraint on gas turbine deployment is not technology it is equipment lead time. Wood Mackenzie estimated in Q2 2025 that lead times for certain turbine classes reached 243 weeks. INNIO’s Jenbacher J920 FleXtra engines, deployed at a 104 MW power plant in Greenville, Texas, reached full load within two minutes of startup a capability directly relevant to data center load management — but the manufacturing queue for those engines stretched well into 2027. Caterpillar, INNIO, and Wärtsilä have all reported record order volumes from data center power projects. INNIO described its 2.3 GW order from VoltaGrid as the largest in company history, delivered across 92 power packs each producing 25 MW. That single order reflects the scale at which modular gas generation is now being deployed in the data center market.
Carbon Capture on Gas Turbines: Engineering Integration
Carbon capture integration on behind-the-meter gas generation is advancing from an aspirational technology add-on toward an operational requirement for projects in jurisdictions with binding emissions commitments. The engineering approach involves post-combustion capture systems attached to turbine exhaust streams, with CO₂ compressed, transported, and sequestered or utilized commercially. Ramboll’s engineering teams working on data center microgrid projects note that carbon capture is applied to both gas turbines and fuel cells in configurations where clients have committed to net-zero operation targets. Captured CO₂ from fuel cell installations has found commercial application in food-grade manufacturing — carbonated beverage production — and in agricultural and industrial processes that consume CO₂ as a feedstock. These circular-economy applications for captured carbon reduce the net cost of capture by creating a revenue stream from the waste product.
Natural gas combined-cycle gas turbines with Selective Catalytic Reduction and low-NOₓ burners, paired with carbon capture and sequestration systems, represent the highest-performance near-term configuration for emissions-constrained behind-the-meter gas generation. Orrick’s infrastructure guide notes that this configuration positions the gas generation asset for future hydrogen blending — transitioning fuel input progressively toward green hydrogen as electrolysis and hydrogen transport infrastructure develops, without replacing the physical turbine. That transition pathway matters commercially: a gas turbine installed today with hydrogen readiness retains asset value through the decade-long energy transition rather than becoming stranded infrastructure when carbon regulations tighten.
Co-Investment in Independent Power Assets
Power purchase agreements were the data center industry’s primary energy procurement mechanism through the early 2020s. A hyperscaler signed a contract with a renewable energy developer, agreed to purchase output at a fixed price over a defined term, and reported the contracted capacity toward its renewable energy targets. This model served sustainability accounting requirements. It did not solve the physical power delivery problem. A PPA delivers contractual rights to electricity — it does not accelerate the construction of the transmission infrastructure needed to move that electricity from a wind farm in West Texas to a data center in Ashburn, Virginia. Operators facing multi-year grid interconnection queues needed a different mechanism, one that placed generation assets physically adjacent to the facility rather than connected through congested transmission networks. Direct co-investment in independent power assets represents that different mechanism.
That structure — equity stakes in co-located generation assets rather than contractual power purchase rights — reflects a fundamental change in how data center operators relate to their power supply. Google’s June 2025 collaboration with CTC Global to deploy advanced high-performance conductors across the U.S. grid was not a PPA. It was a direct co-investment in physical grid infrastructure that increases the capacity of existing transmission lines. Microsoft has surpassed Amazon as the largest buyer of clean power globally, with over 34.7 GW contracted as of late 2025. Their contracts increasingly span solar farms, wind projects, battery storage, nuclear plants, and natural gas generation — a portfolio approach rather than a single-technology procurement strategy.
Wind Farms and the Direct Ownership Model
Hyperscalers’ approach to wind energy procurement has shifted from financial instruments to asset ownership. In the 2021–2024 period, Google executed wind PPAs in the 90–150 MW range through intermediaries. By early 2026, single transactions had reached gigawatt scale — Google’s 1 GW solar PPA with Total Energies in Texas set a new benchmark, structured around two specific projects: the 805 MWp Wichita farm and 195 MWp Mustang Creek facility. These are not abstract renewable energy credits attached to a distant asset. They represent power from named physical assets feeding the ERCOT grid where Google’s Texas data centers operate. S&P Global’s Sustainable1 analysis confirmed that U.S. data centers have contracted more than 80 GW of clean energy to date — roughly 30 GW above estimates made in 2024 — with that procurement representing more than half of total U.S. corporate clean energy contracting.
FTI Consulting’s 2025 M&A review documented a distinct pattern emerging among the largest hyperscalers: moving from financial energy hedges toward direct asset ownership to secure reliability. This manifests in two forms. The first is acquiring operating generation assets with established grid interconnection rights — particularly vintage solar farms that can be retrofitted with battery storage, bypassing the frozen interconnection queue entirely by using an existing point of interconnection. The second is funding the development of next-generation generation assets — SMR projects, advanced geothermal, green hydrogen — through equity commitments that give the hyperscaler a priority offtake claim on the capacity, rather than competing in open markets for power from assets built without a committed buyer.
Utility-Scale Battery Storage in the Microgrid Stack
Battery energy storage systems have entered the data center power stack not as backup UPS replacements but as active grid-management assets. The distinction is operationally significant. A traditional UPS battery provides bridging power for minutes during a utility outage, sufficient to allow diesel generators to reach operating speed. A utility-scale BESS integrated into a data center’s behind-the-meter microgrid provides frequency response in milliseconds, smooths the demand transients that GPU training clusters generate, participates in day-ahead energy markets to optimize fuel costs, and can dispatch stored energy to the grid during peak demand periods when the operator has contracted for export capacity.
U.S. operating storage capacity reached 37.4 GW by October 2025, growing rapidly with another 19 GW under construction through 2026. Deloitte’s 2026 Renewable Energy Industry Outlook notes that over half of utility-scale storage coming online by 2026 is paired with solar, concentrated in southwestern states. For data center microgrids specifically, battery storage serves a different function than in solar-paired grid assets. The storage must handle the specific demand volatility profile of AI compute — abrupt ramps, sustained peaks, rapid drops — while also providing the grid services that regulators increasingly require from large behind-the-meter facilities as a condition of permitting or interconnection flexibility. That combination of AI workload management and grid service participation is what makes battery storage a structurally necessary component of the hypergrid architecture.
Storage Procurement and the Interconnection Bypass Strategy
FTI Consulting identified a specific M&A pattern in 2025 that illustrates how battery storage interacts with interconnection constraints. Independent power producers acquired vintage solar assets — projects with established grid connection rights from years-prior interconnection agreements — specifically to retrofit them with battery storage. The retrofit adds storage capacity to an existing point of interconnection without triggering a new interconnection study. That bypass of the frozen interconnection queue has made brownfield storage retrofits one of the most valuable development strategies in constrained markets. A data center developer co-investing in such a project secures grid-connected generation and storage capacity without entering the multi-year queue that a new interconnection request would require.
Energy Vault and Peak Energy announced a 1.5 GWh sodium-ion storage supply agreement in February 2026 for AI data centers — a deal that signals alternative battery chemistries entering the data center power market alongside established lithium-ion systems. Sodium-ion technology’s supply chain differs from lithium-ion in ways that reduce exposure to the geopolitical sourcing constraints affecting lithium and cobalt. For data center operators pursuing behind-the-meter independence from both the utility grid and commodity metal supply chains, technology diversification in storage chemistry has strategic value beyond the technical performance characteristics of individual systems.
Operators as Active Grid Stakeholders
The phrase “active grid stakeholder” has moved from aspirational language to regulatory obligation in key U.S. markets. PJM Interconnection, which serves 67 million people across 13 states and the District of Columbia, spent 2025 in an accelerated stakeholder process specifically addressing how to integrate large data center loads without compromising grid reliability. The January 2026 Board determination introduced a “Bring Your Own Generation” expedited track — a formal regulatory pathway that allows data centers to connect to the grid on an accelerated timeline, provided they commit to providing their own generation capacity rather than drawing that capacity from PJM’s shared pool. That provision transforms on-site generation from an operational preference into a regulatory tool with direct interconnection acceleration value.
FERC’s December 2025 order directed PJM to establish transparent tariff rules for co-located data center loads with on-site generation. The order addressed a specific gap: PJM’s existing Behind-the-Meter Generation rules had not kept pace with the scale and complexity of modern data center microgrid configurations. The compliance filing submitted by PJM in February 2026 proposed uniform criteria for what qualifies as retail behind-the-meter generation, a materiality threshold, and revised interconnection requirements for large campus facilities. These regulatory developments effectively formalize the data center’s role as a grid participant rather than a passive load — a status that carries both obligations and opportunities. Operators who provide their own capacity reduce PJM’s reliability backstop burden, which in turn qualifies them for the expedited interconnection process that passive loads cannot access.
Grid Services as Revenue and Obligation
The PJM capacity market has experienced price escalation driven directly by data center load growth. Capacity prices increased dramatically between the 2024–2025 and 2026–2027 delivery years. Ratepayer advocates, including a bipartisan coalition of all 13 PJM state governors, issued a joint Statement of Principles in January 2026 demanding that data centers bear the infrastructure costs of their own load growth rather than shifting those costs to utility ratepayers. That political alignment between state governors of both parties on data center cost responsibility created the regulatory environment in which the “Bring Your Own Generation” framework could advance at the speed it did through the PJM stakeholder process.
Data centers that operate behind-the-meter generation assets in PJM are positioned to participate in demand response programs — agreeing to curtail load during constrained grid periods in exchange for capacity payments. PJM’s complementary Issue Charge exploring customer flexibility solutions explicitly identified three mechanisms for large loads to provide grid support during transmission upgrade construction: voluntary curtailment, energy storage provision, and traditional backup generation. Each of these mechanisms turns a data center’s on-site power infrastructure into an asset that provides value to the grid rather than simply consuming it. The financial value of those grid service revenues can offset a portion of the capital cost of the behind-the-meter generation assets themselves, improving the economics of the investment case for on-site power infrastructure.
The Williams and Midstream Gas Pivot
Williams Companies’ strategic pivot illustrates how behind-the-meter demand is reshaping the entire energy supply chain behind data centers, not just the generation assets themselves. Williams committed over $5.1 billion to a “power innovation” business model that includes Project Socrates, a $1.6 billion initiative. Midstream gas companies historically transported fuel through pipelines and charged transmission fees. The behind-the-meter model changes that relationship fundamentally: Williams is now building and operating dedicated on-site power plants at data center campuses, supplying both the natural gas fuel and the electricity generation as a single integrated service.
This “wellhead-to-plug” model — delivering power from the gas source directly to the data center’s compute infrastructure without the transmission grid as an intermediary — addresses the speed-to-power problem through vertical integration. A midstream company that controls pipeline access, gas storage, and on-site generation equipment eliminates each of the supply chain vulnerabilities that slow conventional behind-the-meter project development. It also positions the midstream company as a long-term power service provider rather than a commodity transport business, locking in contracted revenue streams from data center operators who need guaranteed fuel supply and generation capacity for decade-long facility operational lives.
Energy Transfer and the Direct Supply Model
Energy Transfer’s direct gas supply agreement with a data center operator in Texas represents a parallel business model to Williams’ integrated power approach. Rather than constructing and operating generation assets, Energy Transfer supplies gas directly to the data center’s on-site generators under a firm supply contract that guarantees availability independent of spot market conditions. VoltaGrid’s 2.3 GW deployment for Oracle is backstopped by Energy Transfer’s 140,000-mile pipeline and storage network, providing firm fuel supply for INNIO Jenbacher reciprocating engines that can reach full load within two minutes. The combination of contracted fuel supply, modular reciprocating generation equipment, and ABB automation systems creates a behind-the-meter power platform that can be replicated across multiple sites using the same equipment and supply chain relationships.
Williams, Energy Transfer, Kinder Morgan, and Chevron — through its partnership with GE Vernova — have each committed to data center power as a core business line rather than an opportunistic adjacent revenue source. Chevron’s joint development with GE Vernova, announced in early 2025, aims to establish the first multi-gigawatt co-located power plant and data center targeting Southeast, Midwest, and West region colocation facilities using seven GE Vernova 7HA natural gas turbines. The scale of these commitments from infrastructure companies with decades of capital discipline signals that behind-the-meter data center power is not a speculative technology play. It is an established commercial market receiving the same investment scrutiny that oil and gas infrastructure projects attract.
Europe’s Behind-the-Meter Responses
Europe’s data center markets face the same interconnection constraint as the U.S. market, with some regional variations in regulatory response. Pure DC’s Dublin campus features what the company describes as Europe’s first 110 MW microgrid, designed specifically to accelerate AI-ready infrastructure deployment while bypassing the grid constraints that have effectively paused new utility-connected data center development in the Dublin market. Ireland’s grid operator placed a moratorium on new large-load data center connections in the Dublin region that drew significant industry attention — but similar constraints have emerged through utility mechanisms rather than formal moratoria in Germany, the Netherlands, and parts of the United Kingdom.
West London’s situation involves distribution network constraints that cannot be resolved within any commercially relevant timeframe without fundamental transmission infrastructure investment. Microsoft’s €3.2 billion investment in German data center infrastructure reflects both the demand for European AI compute capacity and the recognition that securing power in European markets requires direct investment in energy infrastructure rather than passive grid connection. The strategic logic differs slightly from the U.S. market — European renewable energy integration creates different grid dynamics than the U.S. thermal generation fleet — but the conclusion is the same: operators who wait for utility capacity to become available in primary European markets will wait for years. Those who bring their own generation move ahead of that queue.
Pure DC, AVK, and European Microgrid Engineering
AVK-SEG’s European microgrid deployments for data center clients illustrate the engineering approach that has gained traction in constrained markets. The company’s joint research with Wärtsilä, published in September 2025, demonstrated that microgrids enable data centers to become active participants in grid stability rather than purely consumptive facilities. AVK secured an exclusive partnership with Rolls-Royce for mtu generation sets across the UK and Ireland, establishing the supply chain foundation for rapid-deployment microgrids in markets where grid access is acutely constrained. The company’s strategy targets tier-2 European markets where grid constraints are most acute — a deliberate choice to address demand that primary market bottlenecks cannot serve.
Eaton’s contribution to European data center microgrid architecture focuses on grid-to-chip power architecture and subsynchronous oscillation detection — technical capabilities that become essential when a data center’s power system is not buffered by the inertial stability of a large utility grid. Subsynchronous oscillation is a grid stability phenomenon that can occur when power electronics-dominated systems — fuel cells, battery inverters, and variable-frequency drives — interact with each other’s control systems in ways that create resonance. Detecting and suppressing that resonance in a large behind-the-meter microgrid requires active monitoring hardware and control algorithms that are not necessary in a conventional grid-connected facility. These are infrastructure engineering requirements that European colocation operators are beginning to specify alongside traditional UPS and backup generation when they build AI-optimized facilities.
The Regulatory Frontier: FERC, BYOG, and the Ratepayer Question
PJM’s “Bring Your Own Generation” track, embedded in the January 2026 Board determination, is the most concrete regulatory development in the data center power debate to date. The framework creates an expedited interconnection pathway for data center operators willing to commit on-site generation capacity proportional to their load. A large load — defined as an individual addition of 50 MW or more at a single point of interconnection — that enters through the BYOG track avoids the standard interconnection queue timeline in exchange for the operational and financial commitment of building and operating on-site generation. That trade-off is commercially attractive for developers who intended to build behind-the-meter generation anyway. It is commercially unattractive for operators who wanted grid power without the complexity of owning generation assets.
The White House National Energy Dominance Council and the bipartisan coalition of all 13 PJM state governors made the political dimension of this framework explicit in January 2026. Their joint Statement of Principles demanded that data centers bear the infrastructure costs of their own load growth rather than allocating those costs to utility ratepayers. That statement created the political cover for PJM’s BYOG framework to advance through its regulatory process without the stakeholder opposition that might otherwise have slowed or blocked it. Data centers that oppose BYOG requirements face the political liability of arguing, in effect, that residential ratepayers should subsidize AI infrastructure costs — a position that carries obvious reputational and regulatory risk.
The DOE and FERC Jurisdictional Questions
The U.S. Department of Energy’s October 2025 directive to FERC proposed federal jurisdiction over new loads above 20 megawatts. The goal was to standardize procedures and create faster approval pathways for projects willing to accept operational flexibility — including occasional curtailment during grid stress events. That proposal reflected the gap FERC’s existing generation interconnection reforms had not addressed: FERC Order 2023’s first-ready, first-served cluster study process applied to generation interconnection but not to large load interconnection. Data centers requesting utility connections above 20 MW had no federal standardization framework and faced varying treatment across different state utility regulatory regimes.
The jurisdictional boundary between state and federal authority over large load interconnection remains unsettled as of mid-2026, with PJM’s compliance filing to FERC pending Commission response. The practical consequence is that data center developers planning behind-the-meter projects in PJM must design power architecture under regulatory uncertainty about how co-located generation will be treated for tariff purposes going forward. PJM’s compliance filing defined uniform criteria for retail behind-the-meter generation, capped at a cumulative nameplate rating of 50 MW or less, with different rules for larger configurations. That threshold and the conditions attached to it will shape how hyperscale campus power architecture is structured in the PJM territory for the remainder of the decade.
Emissions, Community Politics, and the Clean Energy Pressure
Natural gas behind-the-meter generation delivers the speed-to-power advantage that AI deployment timelines require. Gas turbines and reciprocating engines can be procured and commissioned in under 18 months, against utility interconnection queues that extend to seven or more years. That advantage is real and commercially decisive. The limitation is equally real: natural gas generation with a 15–20 year operational lifespan creates emissions commitments that conflict with the net-zero targets hyperscalers have publicly announced. Proposals for natural gas data center power capacity in the U.S. tripled in 2025 compared to the year before. That acceleration has drawn attention from environmental regulators, community groups, and the ESG-oriented institutional investors who provide a significant share of data center development capital.
Fuel cells address part of this tension. Oracle’s revised Project Jupiter design specifically cited a 92 percent reduction in NOₓ emissions compared to the gas turbines it replaced. Water consumption falls to negligible levels compared to conventional thermal generation. Community opposition that formed against the original gas turbine design dissolved when the fuel cell microgrid was announced. The clean energy pressure on data center power decisions is not purely regulatory. It reflects the political economy of large-scale AI infrastructure in communities that are weighing economic development against energy cost impacts on local ratepayers and environmental effects on air and water quality. Fuel cells, combined with appropriate carbon capture on remaining CO₂ output, represent the near-term configuration that best navigates that political economy in regulatory environments where combustion-based generation faces significant permitting headwinds.
Renewables, Storage, and the 24/7 Clean Energy Target
Hyperscalers’ 24/7 carbon-free energy commitment — the goal of matching every hour of consumption with carbon-free generation rather than annual averages — requires firm baseload power that solar and wind alone cannot provide. Battery storage bridges intermittency gaps for hours; it cannot substitute for the continuous output of a baseload power plant across multiple days of low wind and solar irradiance. The clean energy portfolio strategy that emerges from this constraint combines renewable generation with firm baseload — either nuclear, fuel cells, enhanced geothermal, or natural gas with carbon capture — to cover the hours when renewables cannot. Microsoft, Amazon, Google, and Meta have all contracted across multiple technologies to build towards 24/7 CFE portfolios. Deloitte’s analysis places battery storage as the fastest current bridge to 24/7 clean power while firm baseload options develop.
The solar-plus-storage combination already outcompetes natural gas combined cycle in many U.S. regions, making it a cost-competitive foundation for the renewable portion of a 24/7 clean energy stack. That portfolio approach requires co-investment in multiple asset classes simultaneously: wind PPAs in high-capacity-factor regions, solar-plus-storage in southwestern markets, fuel cell deployment for immediate on-site baseload, and nuclear offtake agreements to provide the decade-long firm generation that makes 24/7 CFE mathematically achievable. Each of these asset classes involves direct capital commitment rather than passive energy procurement — a structural shift that has repositioned hyperscalers from energy consumers to infrastructure co-developers across the entire clean energy supply chain.
The Investor View: Capital Is Chasing the Power Stack
Institutional capital has recognized that the data center industry’s growth constraint is power infrastructure, and has repositioned accordingly. KKR and Energy Capital Partners’ $50 billion partnership targeting data center plus generation co-development is the largest single expression of this thesis. Blackstone’s BXDC IPO in 2026 underwrote hyperscaler credit at a 5.75–7.0 percent gross asset yield, reflecting both the scale and the risk profile of data center power infrastructure as an investable asset class. VoltaGrid received a $1 billion strategic equity investment from Blackstone and Halliburton in May 2026 — a combination that pairs infrastructure capital with oil and gas operational expertise, directly relevant to the behind-the-meter gas generation platform VoltaGrid operates.
DigitalBridge, which crossed $119 billion of assets under management by March 2026, operates multiple data center platforms globally alongside dedicated power infrastructure funds. Stonepeak launched its Montera Infrastructure platform in 2025 with $1.5 billion in equity — a fourth North American data center investment vehicle explicitly designed to include power infrastructure co-development. These capital structures represent a fundamental change in how data center infrastructure is financed: the power generation asset and the data center building are increasingly viewed as a single integrated investment rather than separately owned components. That integration changes the capital structure of data center development, the risk allocation between operators and investors, and the governance requirements for behind-the-meter generation assets that must perform reliably to protect the data center investment sitting behind them.
Power Infrastructure as Underwriting Criterion
The shift from passive energy procurement to active power infrastructure ownership has changed how financial underwriters evaluate data center investments. A facility with a signed fuel cell deployment agreement from Bloom Energy, a firm gas supply contract from Williams or Energy Transfer, and a grid service participation agreement with PJM is a fundamentally different credit than a facility waiting for a utility interconnection that may or may not materialize within the deal’s investment horizon. CoreSite’s 2025 State of the Data Center Report and Bloom Energy’s 2026 Data Center Power Report both confirm that data center decision-makers now view on-site generation as a primary infrastructure component, not a backup or supplementary asset.
Orrick’s infrastructure guide notes that behind-the-meter generation dramatically changes the risk profile of a data center investment: the risk shifts from utility delivery timeline uncertainty to permitting, construction, and operational performance of the on-site generation asset — a risk profile that infrastructure capital has deep experience pricing and managing. The political and regulatory environment reinforces this capital reorientation. A bipartisan coalition of state governors demanding that data centers bear their own infrastructure costs is a durable political alignment. The BYOG framework at PJM, FERC’s co-location rules, and the DOE’s push for federal large-load interconnection jurisdiction all point toward a regulatory future in which behind-the-meter generation transitions from a voluntary strategy to a baseline expectation for large data center facilities in constrained markets. Investors who underwrite data center projects without secured on-site power face an increasing probability that their asset will face regulatory obstacles that grid-independent competitors will not encounter.
A Power Architecture the Grid Did Not Design For
The utility substation was not designed for the data center industry of 2026. It was designed for industrial economies with gradual, predictable load growth and centralized generation assets that expanded at the pace of regulated utility planning cycles. Multi-gigawatt AI campuses arriving simultaneously across a dozen major markets represent a demand shock that no planning cycle anticipated, and no regulatory framework was designed to accommodate at the required velocity. The response has not been to wait for utilities to adapt — it has been to construct a parallel power infrastructure that operates alongside the utility grid while bypassing its constraints.
Oracle’s decision to replace gas turbines and diesel generators with a 2.45 GW fuel cell microgrid at Project Jupiter is the clearest single expression of this transition. It is not a temporary workaround pending grid access. It is an architectural choice that delivers primary power to a major AI data center campus through an on-site generation system that eliminates the interconnection queue, reduces emissions materially compared to combustion alternatives, and provides the load-following capability that GPU workloads specifically require. Williams and Energy Transfer are building dedicated power infrastructure to fuel those systems. Brookfield, KKR, and Blackstone are financing them. FERC and PJM are writing the regulatory rules that govern how they interact with the grid as active stakeholders rather than passive consumers.
