Everything About AI
NextFin News - EQT is committing $2 billion to a push into small-scale battery systems for U.S. data centers, betting that the fastest way to power the artificial-intelligence boom is not to wait for the grid but to bring storage and generation directly to the server racks. The move, announced September 18, 2026, puts the Swedish private-capital giant at the center of a widening race among infrastructure investors to solve the single biggest bottleneck facing AI: electricity.
The deal is a direct response to a market that has outgrown the public grid. Data-center power demand is forecast to surge 175% by the end of the decade, the equivalent of adding another top-10 power-consuming nation, according to Goldman Sachs research. Meanwhile, connecting new utility-scale projects to the grid can take five to eight years, an analysis of the AI power buildout found. Small batteries, deployed behind the meter and paired with on-site generation, can be energized in months. For a hyperscaler racing to install AI clusters, that gap is the difference between capturing the AI wave and watching it pass.
Data centers already account for roughly 4% of total U.S. electricity load as of 2023; the Electric Power Research Institute estimates that share could climb to as much as 9% by 2030. Doubling the sector's share of national generation in six years is not something a grid built for flat demand can absorb without new architecture. The architecture that is emerging is decentralized, and EQT is placing a $2 billion bet on it.
The Bet: Small Batteries as the Fast Lane Around the Grid
EQT's $2 billion commitment targets distributed, small-scale battery energy storage rather than the giant grid-connected projects that have dominated the sector. The strategy reflects a structural shift in how data centers are powered: instead of queuing for transmission capacity that may not arrive for years, operators are building their own power infrastructure on site.
The scale of the opportunity is substantial. S&P Global estimates roughly 30 gigawatts of on-site power capacity will be needed by 2030 - about a quarter of all new data-center power demand. That is a market large enough to support dedicated investment vehicles, and EQT is positioning itself to capture it through a portfolio that already spans the energy-to-data-center value chain.
The firm's holdings include EdgeConneX, which builds and operates data centers for hyperscale customers; Scale Microgrids, which designs and operates on-site power systems combining solar, batteries, gas generators and fuel cells; Cypress Creek, Zelestra and OX2, which develop renewable energy projects; and Zayo, which provides fiber connectivity. In 2024, EQT added ju:niz Energy, a German battery-storage developer, and launched its EQT Transition Infrastructure strategy to tie these pieces together.
"One way to address the strain is to actually develop your own generation on site," says Matthew Kestenbaum, a managing director at EQT. "Scale does exactly that."
EQT is no stranger to this space. Its EQT Infrastructure VI fund acquired Madison Energy Investments, a Virginia-based developer of distributed solar and storage, giving it a platform with more than 386 megawatts of operating capacity across the United States. The firm also agreed on March 2, 2026, to join Global Infrastructure Partners, CalPERS and the Qatar Investment Authority in a roughly $33.4 billion takeover of AES, a move that adds utility-scale generation and storage to its toolkit. As of March 2026, EQT reported €269 billion in total assets under management, including €142 billion in fee-generating assets.
For comparison, the total U.S. battery-storage market added roughly 10 to 15 gigawatts of new capacity per year in the mid-2020s, depending on interconnection progress. A $2 billion pool aimed specifically at the data-center niche is large enough to move multiple projects from development to operation within a single investment cycle, and small enough to stay focused on the highest-margin, fastest-to-deploy assets.
Why the Grid Lost the Race
The economics are straightforward: AI cannot wait. A data center built for frontier AI workloads consumes power at a density that traditional facilities never approached, and the sites with the best fiber, land and cooling are often the ones with the most congested grids. PJM Interconnection, the wholesale power market stretching from Chicago to New Jersey, faces the greatest data-center demand growth of any U.S. region - yet it has placed few orders for the large gas turbines needed to meet that demand, constrained by price caps and interconnection delays.
The queue problem is the mechanism. Connecting a utility-scale project to the grid now takes five years or more, up from less than two years a decade ago, according to Lawrence Berkeley National Laboratory interconnection data cited in EQT's research. For a developer, that is not a delay; it is a business-model breaker. A data-center tenant will not sign a lease for a facility that may not be energized for half a decade. The customer who can promise power in 12 to 18 months wins the deal, even at a premium price.
This is why "bridge power" has become an industry term. Operators are installing smaller, faster-to-deploy assets - natural-gas generators, battery arrays, fuel cells - to energize a campus while larger, slower projects catch up. Batteries play a dual role: they smooth the output of intermittent renewables so a data center can claim clean power, and they provide the instantaneous response that AI clusters need when workloads spike.
The mismatch is not limited to PJM. Regional grids across the United States report interconnection queues measured in years, and turbine manufacturers have said large gas units are backordered for several years. Smaller generators face shorter waits, which is precisely why the small-battery and bridge-power model works: it substitutes speed for scale. A 100-megawatt battery array sited next to an existing substation can be live far faster than a 1,000-megawatt gas plant waiting on a new transmission line.
The Cyclical Wave Riding a Structural Shift
It matters to separate the two forces at work here. The immediate surge in data-center power demand is cyclical in the sense that it is tied to an investment boom: AI capital spending is front-loaded, and if model training becomes more efficient or hyperscaler budgets slow, the urgency eases. EQT's own research notes that training AI models is becoming more efficient, so a workload in 2027 could consume a fraction of the energy it does today. That is the mean-reverting leg.
But the grid bottleneck is structural. Transmission buildout is slow everywhere, permitting is hard, and the mismatch between where power is generated and where data centers sit is a feature of the system, not a temporary glitch. Even if AI demand growth moderates, the queue problem does not disappear - it simply becomes someone else's problem. Small-scale, behind-the-meter power solves a structural constraint, which is why the investment thesis survives a cyclical downturn in AI spending.
EQT's own portfolio illustrates the pivot. Scale Microgrids went from developing $5 million to $150 million power systems for universities and commercial customers to completing a single data-center deal that tripled its revenue. In June 2025, Scale surpassed $1 billion in total project financing after closing a $275 million round, funding 140 megawatts of microgrids, community solar and storage across New York, Pennsylvania, California and other states.
"I don't think we appreciated just how big it would become," says Ryan Goodman, who founded Scale in 2016.
The speed of the shift caught even insiders off guard. By late 2024, AI-driven data-center construction was running far beyond what Scale and EQT had anticipated. The market wanted billion-dollar solutions, not the company's typical projects. EQT responded by facilitating Scale's acquisition of Reload, a platform that develops gigawatt-scale data-center campuses paired with rapidly deployable on-site power.
"The Reload transaction definitely wouldn't have happened without EQT," Goodman says. "Scale could have built the capability organically, but it would have taken a lot more time. And this space is moving so fast that speed matters."
The Counter-Thesis: Is On-Site Power a Dead End?
The strongest case against this trade is that behind-the-meter power is an expensive detour that the grid will eventually make obsolete. Critics argue that once transmission catches up and large gas turbines finally get ordered, utility-scale power will undercut on-site generation on cost, leaving small battery projects stranded or forced into low-margin grid-services markets. There is also a fuel problem: much of today's on-site data-center power runs on natural gas, which sits awkwardly with the clean-energy pledges many hyperscalers have signed.
That argument has force, but it mistakes the sequencing. On-site power is not competing to be the cheapest source of electricity over a 30-year horizon; it is competing to be the only source available on a 12-to-18-month horizon. A data center that cannot be energized has a capacity factor of zero, no matter how cheap the eventual grid power would have been. The premium that operators pay for speed is not a markup; it is the price of time.
The gas question is real but manageable. Batteries paired with solar and wind can firm renewable output, and fuel cells offer a lower-carbon on-site option. The portfolio that wins is the one that can blend technologies rather than betting on a single fuel. EQT's spread across batteries, solar, gas, fiber and data-center operations is a hedge against picking the wrong winner.
The signal that would prove the bullish view wrong is specific: if PJM and other major grids clear their interconnection queues and bring new large-scale generation online fast enough that the median time-to-energize a data-center campus falls back toward grid-connected timelines - under two years for a 500-megawatt facility - the premium for behind-the-meter power collapses. Watch the queue-clearance rate and the average interconnection time reported by the regional transmission organizations. If those numbers compress materially by 2028, the structural thesis is broken. If they stay stuck above four years, on-site power moves from workaround to permanent architecture.
Who Benefits, Who Is Exposed
The winners are the companies that own the fast-to-deploy assets and the sites where they can be placed. Battery storage developers, microgrid integrators, fuel-cell makers and on-site gas providers all gain pricing power as long as the queue remains long. Data-center operators with their own power can lease capacity at a premium and win tenants that grid-dependent rivals cannot serve. Investors with platforms that span the whole chain - generation, storage, connectivity and the data center itself - can package integrated solutions that a single-technology player cannot match.
The exposed are the utilities and transmission owners that assumed they would remain the sole gateway to power. If large customers can leave the grid or never join it, the traditional cost-recovery model comes under pressure. So do developers who are stuck in interconnection queues with no on-site alternative: their projects risk becoming unleaseable. And pure-play grid-connected battery projects that depend entirely on wholesale market arbitrage may find themselves competing against better-capitalized integrated platforms.
There is a second-order risk for the broader market. If data centers increasingly generate their own power, the fixed costs of the shared grid do not disappear - they are spread across a smaller base of remaining customers. That pushes retail rates higher, which in turn makes on-site power look even more attractive. The feedback loop can accelerate the very decentralization that utilities fear, and it is one reason regulators in several states have begun scrutinizing large load interconnection rules and cost allocation.
Outlook: Three Horizons, Three Scenarios
Short term, the trade is about speed and permitting. Any project that can be energized before 2028 commands a premium, and the financing windows for battery and microgrid developers should stay open. The risk here is execution: supply chains for batteries and gas generators are themselves constrained, and a project that slips loses its edge.
Medium term, the trade is about contracts. The winners will be those who lock in long-term power purchase agreements with hyperscalers at prices that reflect scarcity, not just cost. The losers will be the developers who build on speculation and find the AI capital-spending cycle has cooled before their projects are live.
Long term, the trade is about architecture. If on-site power becomes the default for new data centers, the sector structurally re-rates: power becomes a core competency of the data-center business, not a utility bill. If the grid catches up, on-site assets migrate to grid services and the premium fades. The base case is a hybrid system in which large facilities run on a mix of grid power and dedicated on-site generation, with batteries providing the flexibility layer that ties them together.
The scenarios split cleanly. In the upside case, queue times stay long, AI demand keeps compounding, and integrated platforms like EQT's command premium valuations as the toll-road owners of AI power. In the downside case, efficiency gains cut per-token energy use faster than demand grows, queues clear, and on-site power becomes a niche reliability product rather than a growth market. The base case sits between: persistent congestion keeps on-site power valuable, but competition compresses margins toward a steady, infrastructure-like return.
For investors, the takeaway is that the AI trade has a power leg, and the power leg has a speed leg. The companies that can deliver electricity fastest - not just cheapest - will capture the value. EQT's $2 billion bet is a wager that in the race to power AI, time is the scarcest resource of all.
Data as of September 18, 2026.
Explore more exclusive insights at nextfin.ai.
