Data Centers Can’t Wait for the Grid. So They Build Their Own.

This article follows our previous blog, 'Your Trading Technology Wasn't Built for the AI Data Center Boom. Is It Ready?' In it, we wrote about the scale of AI data center demand and the pressure it is putting on grid interconnection queues, trading systems, and the regulatory frameworks trying to keep up.

However, the industry is not waiting for the grid to expand. Across the U.S., a set of improvised, pragmatic, and in some cases genuinely creative solutions has emerged, from sourcing secondhand electrical equipment to parking repurposed jet engines outside data center campuses to investing in miniaturized nuclear reactors that don't yet exist at commercial scale. Some of these approaches will work. Some will create new problems. All of them add complexity for the trading and risk systems that have to account for the power flowing through them.

Here is what is happening in practice.

Why Transformer and Switchgear Shortages Are Delaying Data Centers

Before you can connect a data center to any power source - grid, gas turbine, or nuclear reactor - you need transformers, switchgear, and breakers. And right now, you probably can't get them on the timeline you need.

Before 2020, high-power transformers typically arrived within about two years of an order. That timeline has now stretched to as long as five years.[1] Medium-voltage switchgear averaged 44 weeks lead time in Q2 2025 alone. Circuit breakers have jumped 47% in price since 2021, with medium-voltage switchgear up 50%.[2] As of April 2026, almost half of U.S. data centers planned for this year are expected to be delayed or canceled, with shortages of transformers, switchgear, and batteries among the primary causes.[3]

Reuters reported 2025 market shortages of 40% for high-voltage switchgear, 25% for breakers, and 20% for medium-voltage switchgear.[4] The executive branch has called U.S. capacity to produce and deploy transformers and high-voltage transmission components "dangerously limited," invoking Defense Production Act authorities around critical domestic energy capabilities.[4]

The bottleneck is not funding, permitting, or even grid capacity in every case. It is the unglamorous, heavy, slow-to-manufacture equipment that sits between any power source and a data center's servers. You can secure the land, the gas supply, the fiber, and the financing, and still sit idle waiting for a substation component that is backordered past 2028.

How the Secondhand Electrical Equipment Market Is Easing Shortages

When new equipment takes five years to arrive, the secondhand market becomes strategically important. Firms like GridGear have emerged as sourcing partners specifically for this gap, helping project teams find transformers, switchgear, breakers, and major substation equipment when lead times, outages, or project schedules create urgency, connecting surplus and removed electrical gear with qualified buyers across utilities, data centers, engineering, procurement and constructions firms (EPC), oil and gas, and solar developers.[5]

The supply feeding this market is real. Data centers upgrade power infrastructure on five to seven year cycles, and the decommissioned power distribution units (PDUs), remote power panels (RPPs), and switchgear they release are typically well-maintained, fully documented, and in high demand from other operators seeking cost-effective expansion equipment.[6] The same boom that is creating the shortage is simultaneously producing excellent low-hour used inventory as earlier-generation facilities refresh their infrastructure.

Load-bank-tested used generator sets with documented hours ship in weeks, not quarters; a critical advantage when a delayed substation component is the only thing delaying a data center's go-live date.[7]

The secondhand equipment route is not without risk. Refurbished transformers and switchgear require rigorous testing and documentation, and sourcing from the secondary market introduces supply chain complexity that purpose-built procurement does not. But for a project that cannot wait, it is often the only viable path.

Behind the Meter Generation: Data Center Power Without the Grid Queue

The more dramatic response to interconnection delays is to skip the grid queue entirely. Estimates suggest that at least a quarter of incremental data center demand through 2030 will be met by behind-the-meter solutions, on-site generation that bypasses congested public grids and gives hyperscalers the speed and reliability that utilities can no longer guarantee.[8]

The scale of what has already been built is significant. As of mid-2026, approximately 2 GW of behind-the-meter generation capacity is online across four projects. xAI's Colossus 1 and 2 outside Memphis account for 1,498 MW of operating gas turbine capacity, a figure that has grown steadily as xAI added dozens of temporary mobile turbines at the site.[9]

The planned natural gas capacity queued for interconnection between 2026 and 2030 has grown dramatically. At the start of 2025, only 23.8 GW was expected online in this period. By January 2026, this had expanded to approximately 48 GW, and in the first quarter of 2026 alone, an additional 16 GW was added, bringing the total planned pipeline to 64 GW by 2030.[10]

On-site generation takes several forms. Natural gas turbines, both aeroderivative units for rapid deployment and combined-cycle plants for long-term baseload, are the dominant choice. A Chevron subsidiary announced a partnership with GE Vernova to build a new company developing scalable, natural gas-powered solutions for U.S.-based data centers, with the first projects expected to use seven GE Vernova 7HA natural gas turbines serving colocation data centers in the Southeast, Midwest, and West.[11] Some operators have gone further, using repurposed jet engines that once powered commercial aircraft, or deploying mobile turbines on semi-trailers parked adjacent to the campus.

The State of New Mexico blocked a planned gas pipeline that would have fueled one of OpenAI and Oracle's signature Stargate projects, a planned 2.45 GW on-site power plant, and satellite analysis suggests the project is likely to miss its 2027 target as a result.[9] Building your own power is a real option, but it is not immune to the same permitting and regulatory friction that delays grid-connected projects.

Small Modular Reactors: Why Most Nuclear Capacity Is Still Years Away

Further along the innovation spectrum, and further from commercial reality, is the microreactor. Small modular reactors are compact nuclear fission reactors with reported capacity of up to 300 megawatts per unit, designed for factory assembly and scalable deployment. A smaller subset called microreactors, potentially capable of generating up to 50 MW per unit, could support remote operations, microgrids, and future data center development by providing dedicated on-site electricity with built-in redundancy.[12]

The hyperscaler interest is genuine. Oklo and Meta have announced an agreement supporting development of a 1.2 GW nuclear power campus in Pike County, Ohio, to supply clean energy for Meta's regional data centers including its AI supercluster in New Albany. Amazon has backed X-energy, which raised more than $1 billion in an April 2026 IPO, targeting up to 5 GW of new nuclear capacity by 2039.[13] Equinix has pre-ordered 20 microreactors from Radiant for deployment at its urban data center portfolio.[14]

Nuclear offers 24/7 carbon-free baseload power with capacity factors above 90%, positioning SMRs as critical infrastructure specifically because hyperscalers are under shareholder pressure to meet net-zero commitments and can't fill a multi-gigawatt gap with natural-gas peaker plants without breaking those pledges.[15]

Most of this capacity is not expected until 2030 or later. The GE Vernova Hitachi BWRX-300, the furthest along of any Western SMR design, is under construction at Ontario Power Generation's Darlington site in Canada targeting commercial operation around 2030. A single hyperscale AI campus can require 1 to 2 GW, meaning multiple SMR units, not one, would be needed to serve a large data center's full continuous load.[15] Licensing timelines, regulatory frameworks that weren't written for private nuclear installations, and financing structures that don't yet exist at commercial scale are the main obstacles to nuclear power for data centers.

What New Power Arrangements Mean for ETRM, Trading, and Risk Systems

Here is the part that is rarely covered in infrastructure reporting, but matters enormously for the trading and commercial operations teams managing these assets: every one of these power solutions creates a different commercial structure, a different settlement requirement, and a different risk profile.

A utility interconnection agreement is very different from a behind-the-meter gas supply contract from a midstream operator. A bilateral nuclear offtake agreement with a hyperscaler is a different instrument entirely from a standard power purchase agreement. Secondhand equipment sourced through a spot market brings procurement risk, warranty complexity, and asset management requirements that purpose-built infrastructure does not. And a fleet of mobile gas turbines parked outside a data center in Memphis, dispatching on a real-time basis to serve a mission-critical AI workload, requires scheduling, nomination, and settlement infrastructure that many ETRM systems were not designed to support conveniently.

Traders and commercial teams are structuring and managing deals that their platforms cannot fully capture. Settlement teams are reconciling arrangements their tools were never configured to handle. Risk teams are managing exposures their systems cannot adequately model.

The power problem is being solved, or at least worked around, by an industry moving faster than its supply chains, its regulators, and its own software. Back-office teams have yet to deal with the effects of all this improvisation.

In a future article, we will look at what the commercial and operational infrastructure behind these new power arrangements actually needs to look like, and where organizations are seeing opportunity and success.

Get Started with capSpire

capSpire helps energy and commodity trading organizations assess their trading technology environment and select, implement, or optimize ETRM/CTRM systems. Contact capSpire to discuss how well your current technology supports your business.

John Harris
Client Account Manager

¹https://vucense.com/ai-intelligence/industry-business/us-ai-data-center-transformer-shortage-crisis-2026/

²https://www.powermag.com/transformers-in-2026-shortage-scramble-or-self-inflicted-crisis/

³https://prosperousamerica.org/americas-ai-boom-has-a-trade-policy-blind-spot/

⁴https://moonshotus.com/transformer-shortage-and-supply-chain-reality-why-electrical-gear-is-now-the-ai-data-center-bottleneck/

⁵https://gridgearsupply.com/

⁶https://circuitbreakerbuyerusa.com/blog/data-center-electrical-surplus-removal-guide

⁷https://csdieselgenerators.com/the-ai-data-center-boom-and-generator-shortage-lead-times-prices/

⁸https://www.latitudemedia.com/news/behind-the-meter-generation-is-picking-up-traction/

⁹https://cleanview.co/reports/behind-the-meter-data-centers

¹⁰https://www.rbccm.com/en/insights/2026/05/natural-gas-powers-the-data-center-boom

¹¹https://www.coresite.com/blog/more-power-behind-the-meter-power-systems-for-data-centers

¹²https://core.verisk.com/insights/emerging-issues/articles/2026/august/week-5/small-modular-nuclear-reactors-and-data-centers

¹³https://davisgraham.com/news-events/small-modular-reactors-and-data-centers-the-emerging-regulatory-landscape-for-next-generation-nuclear-power-supply/

¹⁴https://www.datacenterdynamics.com/en/analysis/atoms-for-data-small-nuclear-reactors-smr/

¹⁵https://www.coradvisors.net/2026/08/small-modular-reactor-nuclear-power-data-centers-2026.html