U.S. Data Centers Announce 90 GW of Behind-the-Meter Power Plans

Author: Christopher Johnson (Avanza Energy)

Compiled by Deep潮 TechFlow

Shenchao Overview: Amid the current race to secure power at data centers, behind-the-meter self-generation has evolved from a past emergency backup option into the sole lifeline for ensuring project timelines. The 90 gigawatts of off-grid, self-owned power plants already announced in the U.S. signal a fundamental restructuring of computing infrastructure, moving away from traditional public grids.

However, as power generation units and gas supply gradually become standardized, commoditized products, the core competitive focus of the industry has quietly shifted upward. Major electrical giants and energy solution providers are no longer just competing on hardware equipment, but are vying for control of the “system integration and electrical control layer” spanning from power generation to compute chips.

This article provides an in-depth analysis of the latest strategic alliances and business withdrawals by industry giants such as Schneider Electric and Caterpillar, revealing the ultimate determinant in the transition of on-site power supply models from engineering packages to commercial implementation: whoever controls the control logic and single-line diagram architecture will establish the deepest commercial moat over a twenty-year computational lifecycle.

Cleanview’s latest tracking data shows that 59 data center campuses in the U.S. have collectively announced approximately 90 gigawatts (GW) of behind-the-meter self-generated power capacity, with about 82 GW announced intensively since early 2025.

However, the actual behind-the-meter capacity currently in operation is only about 2 gigawatts. [1] This significant gap highlights just how newly emerged this demand is, rather than any instability in the project pipeline.

Before 2025, on-site self-generated power was considered only an emergency backup solution; today, it has become a mandatory requirement for data centers to be powered on schedule. Project timelines have become the most stringent hard constraint: as the scale of under-construction data centers in North America reaches a historic high, traditional grid connection waiting lists are being completely exhausted.

The energy research firm Enverus has provided clear capacity projections: between 2026 and 2030, 25.5 gigawatts of new industrial electricity demand in the U.S. will be met through behind-the-meter self-generation, with data centers accounting for 22.5 gigawatts—approximately 36% of all new data center capacity added in the U.S. during this period. [2]

Meeting this massive demand requires 31.6 gigawatts of new on-site power generation capacity, of which 29.6 gigawatts come from natural gas. Reciprocating internal combustion generators account for 7.1 gigawatts, small and medium-sized gas turbines for 6.2 gigawatts, and fuel cells for 4.8 gigawatts.

Chart: 59 data centers in the U.S. have announced approximately 90 GW of behind-the-meter installed capacity; Enverus forecasts behind-the-meter load to reach 22.5 GW by 2030 (Source: Cleanview and Enverus Intelligence Research)

These figures depict stacks of hardware equipment, but they don't address: when an entire power plant fails to come online on time, who should the customer hold accountable?

The series of announcements by suppliers at the Yotta 2026 conference centers precisely on this core issue—the most compelling aspect of the entire industry: the technological pathways for power generation equipment have largely been settled, but their commercial fulfillment structures remain uncertain.

Avanza has previously systematically mapped out the scale of on-site power expansion, the typical parks behind it, and the five key technology pathways driving its development. However, what has not yet been clearly defined is the critical control layer built on top: who is responsible for integrating all components? Who will own the core customer relationships over the next two decades? And who is actively choosing to exit this layer?

On September 30, Wärtsilä, Schneider Electric, and Stanley Consultants jointly launched an end-to-end power solution for U.S. data centers, called "Generator-to-Chip." [3]

The roles of the three parties are clearly defined: Wärtsilä provides modular gas engine power plants; Schneider Electric handles the electrical architecture, automation systems, and digital energy management; and Stanley Consulting oversees engineering design, permitting, multi-party coordination, construction management, and commissioning supervision.

Schneider has packaged this system into the concept of an "AI Energy Park." Melton Chang, Executive Vice President of Power Systems at Schneider Electric, precisely defines it as: "An AI Energy Park starts with computing load and integrates power generation, electrical infrastructure, automation, and digital technologies into an organic system." [3]

Risto Paldanius, Vice President of Wärtsilä Energy in the Americas, also revealed the original intent: “By integrating our complementary capabilities from the very beginning, we can streamline project execution, reduce handoff losses, and significantly shorten construction timelines.”

Carefully examine these verbs: collaborate, streamline, reduce handoffs. At its core, this is a coordinated action agreement among three independent companies aimed at presenting a unified front. This is certainly not a single-contract turnkey solution, and the announcement does not make such a promise.

The announcement contained no contract structure, no megawatt capacity metrics, no disclosed names of specific customers or pilot sites, and no quotations whatsoever. StorageReview, in its independent reporting on the partnership, noted that the release "provided no numbers, customers, or timelines." [4]

The public materials disclosed by Schneider to date have only covered the technical engineering aspects: the division of labor logic from generators to chips, keynote speeches from the Yotta conference, and the switchgear white paper.

The actual commercial terms—namely, who signs the contract with the end customer and who leads the long-term business relationship—remain entirely absent from public records. If these were verified, contracted clients, their installed capacity would typically be publicly announced or at least disclosed; however, there has been no such announcement to date.

The most logical interpretation is that the entire industry, including Schneider, is moving much faster to engineer and productize on-site power than to business-model it—and the latter is often far more challenging than the former.

For seven years, Schneider owned a physical business specifically designed to address all of the above business challenges.

AlphaStruxure, established in 2019, is a joint venture between Schneider and The Carlyle Group. It designs, finances, builds, owns, operates, and maintains on-site energy infrastructure, offering fixed electricity rates and performance guarantees with contractual assurances.

It undertook the 11.4-megawatt microgrid project for the new Terminal 1 at New York’s JFK Airport, to be built in phases until 2029;[5] and in April 2026, it completed the largest bus depot microgrid in the U.S. in Montgomery County, Maryland, featuring 4.8 megawatts of solar photovoltaic capacity, 2 megawatts of energy storage, and the ability to operate in permanent island mode during grid outages.[6]

However, Schneider no longer describes itself as the parent company. As of April 19, 2026, AlphaStruxure’s official homepage still states, “Our parent companies are Carlyle and Schneider Electric.”

But by May 8, the same passage had been quietly revised to the singular form “our parent company Carlyle,” with Schneider’s name completely erased. The surrounding context remained unchanged word for word, indicating this was an extremely precise and targeted revision—a formulation that remains on their website to this day. [7]

This revision of the webpage copy was not accompanied by any official trading announcement; there is no trace of it in press releases, SEC filings, or industry media reports. Carlyle’s regular financial reports also cannot verify this, as the 10-K filings of asset management firms only disclose management fees and AUM, not the individual companies held by their funds.

But the undeniable objective fact is: Schneider no longer presents itself as the parent company and has withdrawn from the entity that previously directly handled asset ownership and commercial engagements.

The Energy-as-a-Service (EaaS) model is unquestionable. Zero-CapEx on-site power is a vibrant and well-capitalized competitive arena: VoltaGrid raised $1 billion from Blackstone and Halliburton in May 2026, achieving a valuation exceeding $10 billion, backed by its ownership and operation of power plants and direct electricity sales to customers. [8]

Yet, just as on-site power generation became the most fiercely contested battleground across the industry, Schneider voluntarily stepped back from this capital-intensive investment layer.

Looking at it positively—and I believe correctly—this is a strategic decision about which layer is truly worth holding long-term.

What Schneider firmly retains is the electrical control and automation architecture layer, and the company is demonstrating its capital logic with real money: the control layer is the ultimate bottleneck of the entire chain.

The software-defined medium-voltage switchgear unveiled at the Yotta conference is claimed to be the world’s first, achieving complete decoupling of system intelligence from physical hardware. Schneider asserts that, compared to traditional designs, this solution reduces control wiring by 87%, terminal blocks by 91%, and copper usage by 85%, while increasing order-to-manufacture speed by up to three times and cutting on-site commissioning time in half. [9]

The pilot project will run until 2027 and be fully launched to market in 2028. Some components of the technology stack are already being deployed earlier: 2 MW, 2.25 MW, and 2.5 MW prefabricated power modules based on the Galaxy VXL UPS are already available in Europe and are committed to launching in North America by the end of 2026. [10]

The party that leads the development of the one-line diagram and control logic naturally becomes the default leader for equipment commissioning, operations and maintenance services, and system upgrades throughout the asset’s twenty-year lifecycle.

Power generation units will eventually be replaced and phased out, and server racks are fully upgraded every three to five years; however, the underlying electrical topology and relay protection schemes remain enduring, along with the unbreakable commercial ties to the system operations team.

In addition, a purely control system positioning can be horizontally compatible and sold to all projects on the market. Schneider has precedent for this: in March 2024, it partnered with Mainspring Energy to bundle its proprietary control and integration system with third-party linear generator peripherals. [11]

Partnering with Wärtsilä simply follows the same logic by pairing it with a higher-power engine. This same approach can be seamlessly replicated with Caterpillar power plants, Bloom fuel cell units, or power company-invested units. Owning an energy park allows you to capture the revenue from just one park; but by controlling the system integration layer, you can share in the profits of all energy parks.

The strong financial performance fully validates this investment. Schneider reported revenue of €21.2 billion for the first half of 2026, with organic growth of 14.0%, including €11.5 billion in the second quarter, representing organic growth of 16.5%. [12]

The pure data center segment saw triple-digit order growth in the second quarter, while overall data center and network terminal market actual sales maintained strong double-digit growth. The "Systems" business model, which encompasses prefabricated components and EPC services, accounted for 35% of second-quarter revenue and achieved impressive organic growth of 28%.

First, the impressive metrics cited above for the switching equipment come entirely from Schneider's own White Paper No. 303, without any independent third-party audit, verification from core customers such as Equinix, or any publicly disclosed pilot test data. [13]

Until genuine reference customers come forward to endorse them, claims of saving 87% in wiring, 91% in terminal blocks, and 85% in copper, along with multi-fold speed improvements, should be viewed with the usual skepticism applied to vendor marketing statements.

Figure: Schneider Electric's claimed performance metrics for its software-defined medium-voltage switchgear (Source: Schneider Electric, unaudited)

The technology media outlet The Register raised more pointed questions, which Schneider’s public materials have so far avoided addressing: “The catastrophic consequences of executing incorrect or malicious software updates on critical electrical systems that affect human and equipment safety are vastly different from mistakenly applying a desktop patch on an office computer.”[13]

Moving medium-voltage electrical protection and control logic to the software layer introduces efficiency gains but fundamentally entails a significant exchange of operational risk.

A data center that physically disconnected from the public grid to bypass grid connection waiting lists has also lost its final line of safety protection. At this moment, this purely software-based protection system has become the only barrier between sudden load fluctuations and the entire campus plunging into darkness.

The high degree of similarity in naming indicates this is an industry-wide structural repetition, not merely a marketing tactic by individual companies.

As early as June 2025, Eaton and Siemens Energy partnered to bundle a standard 500-megawatt gas turbine unit with an energy storage system, quantifying the commercial return as an additional $2 billion to $3 billion in revenue for a typical 500-megawatt facility by accelerating its commissioning by up to two years. [14]

Vertiv and Caterpillar followed suit by forming an alliance in November 2025, with Vertiv’s CEO Gio Albertazzi explicitly defining it as a “Grid-to-Chip” framework. [15]

Schneider and Wärtsilä will launch in September 2026 under the banner "Generator-to-Chip."

Within just fifteen months, three major industry pairings have rapidly come together, with two of them even independently adopting nearly identical "X-to-chip" messaging frameworks. The entire industry has reached a consensus: the smallest unit of competition is no longer a single engine or switchgear, but the complete pathway from source fuel to end-point computing power. The only unresolved question is who will dominate the sales of this pathway.

Figure: Six bundled integration models for on-site power supply and the role distribution of each party at the integration level (Source: Avanza Energy Analysis)

Meanwhile, two other business models exist entirely outside this bundling race, defining the boundaries of the industry.

One model involves a utility power company owned by investors directly owning power plants, purchasing generating units directly from OEMs, with the OEMs retaining no software layer, and the power company selling electricity directly to data centers.

The other approach involves project developers financing, owning, and operating power plants themselves, selling electricity directly to customers behind the meter—this is precisely the flagship model upon which VoltaGrid is built and the path once taken by AlphaStruxure.

Beneath these five pathways lies the most fundamental competitive baseline for all bundled solutions: customers independently sign three separate procurement contracts and bear full responsibility for integration. This results in six distinctly different configurations for deploying the same megawatt of power, with the integration control layer awarded to a different provider in each model. Only in Schneider’s model is the integration layer firmly placed in the hands of the software provider that writes the control logic.

The evolution of regulatory policy has made this business issue more urgent than ever, with momentum far exceeding what was apparent at the Yotta Conference. On December 18, 2025, the Federal Energy Regulatory Commission (FERC) issued an order under Section 206 of the Federal Power Act, finding that PJM’s grid rules lacked sufficient clarity and consistency in rates and interconnection terms for utilizing new generation resources to serve co-located loads, and imposed a regulatory definition.

PJM submitted its compliance filing on January 20, 2026, and FERC issued a formal order on April 16, 2026, approving some aspects and rejecting others. [16]

The portion adopted by FERC provides a direct and substantial benefit to energy parks: applicants seeking grid connection services for on-site loads may now apply for grid services based on their actual load, which is lower than the generator’s nameplate capacity. PJM must calculate system expansion and upgrade requirements based on the actual declared capacity, not the machine’s maximum theoretical output.

Customers can fully leverage existing acceleration clauses to bypass subsequent complex evaluation stages with minimal grid impact, applying for temporary grid connection to begin power generation and operation before full facility completion, while also claiming surplus capacity service rights at existing connection points.

It is precisely these four specific mechanisms that transform the so-called "grid flexibility" from mere marketing rhetoric into legally binding provisions in electricity pricing regulations; this is also the underlying logic that enables a computing park currently operating as an isolated microgrid to potentially comply with regulations and sell back peak-shaving flexibility to the main grid in the future.

The portions rejected by FERC serve as a stark reminder that regulations are still in flux. FERC firmly rejected PJM’s attempt to substitute the term “point of interconnection” with “point of ownership change” in the definition of on-site load, and dismissed PJM’s proposed modifications to the behind-the-meter self-generation application process on the grounds that they fell outside the scope of this proceeding, ordering PJM to make further corrections to ensure compliance.

Deep-seated contradictions have been fully exposed: PJM’s own Independent Market Monitor (IMM) has issued a warning that the current plan provides no clarity on how the public grid will absorb the massive, sudden influx of reverse power from on-site facilities when large-scale computing loads experience an abrupt collapse—or how the grid system will plan for backup power supply in the event of an unexpected shutdown of on-site generators.

FERC temporarily deferred resolution of these complex issues as beyond the scope of this proceeding. Today, these longstanding challenges have been fully transferred to the massive load docket, and FERC has taken action: on June 18, 2026, it issued Order No. 206 simultaneously to the six major regional transmission organizations (RTOs) and independent system operators (ISOs) nationwide, requiring each to demonstrate that their current rules regarding massive load and behind-the-meter self-generation are just and reasonable—or face comprehensive reform. [17]

Following an extension through August, the final formal hearings for each region concluded in mid-November. This means that reliability standards required for a previously isolated facility to connect to and sell power to the grid are now being hastily drafted across major grid regions, with no nationwide consensus yet established. Avanza previously provided an in-depth analysis of the 60-day countdown window granted by FERC to the six major grids, demonstrating that stringent environmental review permits for projects—not merely amendments to FERC procedures—are the fundamental bottleneck constraining gas-fired generation after the deadline.

Three decisive indicators will reveal the ultimate fate of this business model faster than any official press release:

First, the officially named initial client and the form of their contract. If the final agreement is a single turnkey contract assuming full responsibility, it means that Schneider or its competitor has successfully penetrated the system integration layer and firmly secured customer lock-in.

If the customer continues to sign three separate contracts for equipment and engineering, it indicates that the customer still assumes the role of the final system integrator; the so-called ecosystem bundle is merely a scheduling tool designed to expedite timelines and has not driven genuine business model transformation.

Second, whether the project financing can be successfully secured, and which party’s balance sheet will bear this burden. Schneider has previously participated in syndicated financing; it is worth noting the roles of the participants involved.

In September 2025, Schneider, in collaboration with Microsoft and more than twenty allies, launched the Accelerating Resilient Infrastructure Initiative, with six project developers securing $7.5 billion in dedicated funding to deliver projects through Energy-as-a-Service contracts. Schneider positioned itself explicitly as a strategic integrator, facilitating connections between clients and developers while providing technology—one of the six developers carrying this substantial funding being AlphaStruxure. [18]

Just seven months later, Schneider’s name was completely removed from the list of AlphaStruxure’s parent companies. Regardless of the underlying equity restructuring mechanisms, its strategic direction remained consistent: Schneider firmly retained its position as the integrator, leaving the heavy burden of asset ownership to the balance sheets of other major players.

Major external capital players are accelerating their entry. In June 2026, Brookfield significantly expanded its strategic partnership with Bloom Energy on AI computing power infrastructure by fivefold to $25 billion, specifically dedicated to investing in on-site power infrastructure. [19] Data centers initially entered the behind-the-meter self-generation space simply because it was the fastest available path to secure power; now, capital markets are creating massive funding channels tailored specifically for them—marking a clear milestone in the maturation of this niche sector.

Finally, can the electrical control layer’s betting withstand the test of real-world combat? Schneider’s standardization legacy in this field is far longer than the Yotta conference announcement suggests.

Back in 2023, it launched the EcoStruxure Microgrid Flex standardized microgrid solution, [20] and has continued to expand its channel and financing partners throughout 2025.

In March 2026, it officially inaugurated its microgrid test lab at its Andover global R&D center, with its Global Vice President of Microgrids explicitly stating the goal as “accelerating the deployment of highly robust microgrid control systems through a more standardized approach to developing tested, validated, and well-documented reference architectures.”[21] This indicates that, fully six months before the launch of the AI Energy Park, Schneider had already been openly commercializing its microgrid control architecture.

Chart: Publicly recorded timeline from 2019 to 2028, highlighting key milestones for Schneider Electric and AlphaStruxure

Schneider has never publicly linked its AI energy park directly to its existing microgrid business, as the specific product lines within a multinational corporation are often independent of the early departments that developed those capabilities. However, this clear chronological progression is highly instructive: it provides the strongest evidence that its comprehensive leap into the control layer is a deliberate, well-planned strategy—not a fleeting, opportunistic trend.

Jim Simonelli, Senior Vice President and Chief Technology Officer of Schneider Electric’s Secure Power Business, cut to the objective reality during NVIDIA’s GTC conference in April: “Three years ago, the industry’s limiting challenge was how to remove heat from cabinets; today, the real challenge is how to bring power into cabinets and onto campuses.”[22] When discussing dispatchable on-site power capable of meeting AI-scale demands, he was even more blunt: “Aside from natural gas, there is currently no viable alternative.”

Natural gas is a highly mature, undifferentiated commodity, and internal combustion engines are off-the-shelf industrial products with clearly marked prices. Throughout the entire value chain, only system integration and electrical control layers can continuously generate compounding value through technological expertise.

The supplier who first enables real paying customers to adopt and pay for their integrated solution will officially reveal the true commercial ceiling and ultimate valuation of this critical control layer to the entire industry.