Everything About AI
NextFin News - The chief executive of FuelCell Energy is warning that America's artificial-intelligence lead will be forfeited not by a shortage of chips but by a shortage of power, and that moratoriums on data centers solve nothing because they relocate demand rather than reduce it. Jason Few, president and chief executive of FuelCell Energy (NASDAQ: FCEL), made the argument in a television interview on October 30, 2026, as his company's shares fell 7.88% to $11.22 at the close.
Few's message is a direct challenge to a growing political backlash against AI infrastructure. Communities and state governments are increasingly resisting data-center construction over concerns about electricity bills, water use, and fossil-fuel generation. The answer, Few argues, is not to slow the buildout but to re-engineer how data centers are powered — moving generation behind the meter so that compute no longer waits on the transmission grid.
The Political Backlash Meets a Physical Constraint
The tension Few is navigating is concrete. In Newport Township, Pennsylvania, FuelCell Energy would supply power to a proposed data-center project but is not the developer — yet the company finds itself at the center of a community debate over whether the facility should be built at all. Few's position in the interview was that communities should have a say, but projects should be judged on their actual effects rather than rejected simply because they are data centers.
That local dispute is one front in a national fight. New York Governor Kathy Hochul signed an executive order imposing what was described as the nation's first statewide moratorium on new hyperscale data centers while regulators develop a framework for how large facilities connect to and pay for energy infrastructure. Similar temporary bans and permitting pauses have emerged across the country. A Gallup survey released this spring found that seven out of 10 Americans would oppose a data center being built near them — including nearly half who say they strongly oppose such projects — and more would rather live near a nuclear power plant than a data center. Local opposition, one analysis estimated in July 2026, had already stalled roughly $130 billion in AI data-center projects.
The politics are understandable. Data centers are visible, land-intensive, and power-hungry, and the costs of grid upgrades have historically been socialized across ratepayers. But Few's core argument is that the opposition is aimed at the wrong bottleneck. The constraint is not political permission; it is physics and infrastructure capacity.
The biggest threat to America's AI future is not a lack of innovation. It is the mistaken belief that slowing infrastructure somehow creates better infrastructure.
Few wrote that in an October 5 op-ed. And the demand side of the equation is not discretionary. ChatGPT reached 100 million users in two months, a pace social media platforms took years to match. AI now underpins search, healthcare, logistics, software development, and financial services. Halting data-center construction in one jurisdiction does not reduce that demand; it redirects the capital elsewhere.
The Grid Math Does Not Add Up
The reason the "pause" argument fails on engineering grounds is simple arithmetic. BloombergNEF estimated in July 2026 that U.S. data centers are on track to consume about 20% of the nation's electricity by 2035, up from 5.9% today. The research firm raised its 2035 demand projection to 194 gigawatts — an 83% increase on the forecast it issued in December — and calculated that even a record pace of grid connections sustained every year for a decade would still leave a 19-gigawatt supply shortfall. The most data-center capacity ever connected to the U.S. grid in a single year is 7.1 gigawatts.
The 2035 figure has increased roughly 2.5 times in 15 months, reflecting the volume of AI facilities entering the development pipeline rather than capacity actually under construction. Nearly half of the projected capacity will serve AI training and inference workloads, and the U.S. is expected to host 64% of the world's AI chips by power demand in 2033.
Morgan Stanley reached a similar conclusion on a shorter horizon. The bank raised its projected U.S. data-center power shortfall through 2028 to 33 gigawatts, equal to 34% of the 97 gigawatts of incremental power developers will need. Its model shows the cumulative midpoint shortfall rising from 5 gigawatts in 2026 to 12 gigawatts in 2027 and 33 gigawatts in 2028. Extending the forecast through 2029 produces a 72-gigawatt residual gap even after accelerated power measures.
The grid operator itself is sounding the alarm. PJM Interconnection, the largest U.S. power grid, warned in July 2026 that data centers may face involuntary outages under a plan to avert widespread blackouts and protect residential ratepayers from price spikes. Digital warehouses that fail to secure enough generation could be temporarily dropped from the grid during high-demand periods as soon as mid-2027.
This is the mechanism behind Few's urgency: interconnection queues are long, transmission buildout is slow, and community opposition is lengthening both. A moratorium does not shorten the queue. It simply determines which region loses the investment.
Behind-the-Meter Power as the Structural Answer
Few's proposed solution is architectural rather than regulatory. Fuel cells can generate electricity on site, without combustion, with substantially lower emissions, noise, and water use than conventional combustion-based generation. Paired with battery energy storage, they reduce dependence on constrained transmission and avoid shifting infrastructure costs onto other ratepayers. The technology is quieter than combustion power systems and can be deployed in stages as a data center's load grows.
This is not a hypothetical product line for FuelCell Energy. In January 2026 the company partnered with SDCL to deploy up to 450 megawatts of fuel cells across data centers globally. In July it partnered with AI data-center developer Inuverse to explore up to 100 megawatts at a facility in Daegu, South Korea. In March it launched a modular 12.5-megawatt data-center energy block. And in September it reported signing its first Capacity Reservation Agreement with a major data-center operator for a planned 75-megawatt project in Texas, supported by an upfront reservation payment.
The company is also scaling manufacturing to meet the opportunity. Its Torrington, Connecticut facility is being expanded to an annualized output of up to 500 megawatts, at an estimated cost of $200 million to $275 million, with construction underway over two years. FuelCell Energy's carbonate fuel systems come in 2.5-megawatt and 1.25-megawatt units, can operate on natural gas, biogas, or natural gas blended with up to 50% hydrogen, and can cogenerate hot water, high-pressure steam, or chilled water to raise overall efficiency.
Few frames this as a generational shift in data-center design.
As AI and high-performance computing scale, power is no longer just about more capacity—it's about a different architecture. FuelCell Energy natively generates continuous, megawatt-scale direct DC power behind the meter.
SDCL chief executive Jonathan Maxwell said the technology offers "reliable, high-availability power with significantly lower emissions," adding that its "flexibility and efficiency make it particularly attractive for data centers, where resilience and sustainability increasingly need to go hand in hand."
The Cyclical-versus-Structural Call
The critical question for investors is whether the AI power shortage is cyclical — a temporary squeeze that will revert as grid capacity catches up — or structural, a regime shift that will not self-correct. The evidence points decisively to structural.
A cyclical shortage has three markers: a short-term supply or demand shock, a clear mean-reversion path, and historical precedent. The current situation has none of them. The demand shock is not a single surge but a compounding curve: data-center power demand is forecast to rise roughly fourfold over the coming decade, driven by AI workloads whose adoption is measured in months, not business cycles. The supply response is not delayed by a temporary bottleneck but by interconnection queues that take years, permitting processes that are being tightened rather than loosened, and a transmission grid whose buildout pace — 7.1 gigawatts in a record year — is structurally incapable of closing a 19-gigawatt gap by 2035.
History does not offer a reversion precedent because the underlying driver is new. Previous technology buildouts — fiber optics, cloud computing, mobile networks — did not concentrate gigawatt-scale loads in single campuses with the density of a modern AI training cluster. The load profile itself has changed: a typical business server rack draws 5 to 10 kilowatts, while an AI rack packed with graphics processors routinely demands 40 to 80 kilowatts, and the latest liquid-cooled systems push past 100 kilowatts per rack. That is precisely why operators are moving to onsite generation rather than waiting for utility upgrades.
The second-order implication is where the investment thesis actually lives. The first-order effect of the power shortage is obvious: data centers need more electricity. The second-order effect is that the architecture of data-center power changes permanently. If generation moves behind the meter, the value chain shifts from utilities and transmission developers to distributed-generation equipment makers, fuel suppliers, and balance-of-plant integrators. FuelCell Energy's argument that the industry is moving toward centralized 800-volt DC power inside data centers is part of this re-architecture — its systems generate DC natively and can transition from AC-coupled deployments today to direct DC designs without replacing core modules.
The third-order effect is geographic. If compute can be powered onsite, data centers are no longer anchored to regions with available grid capacity. Capital flows to jurisdictions with predictable permitting and clear energy access — which is why Few warns that development paused in one region moves elsewhere, including to the Middle East and Asia, where governments are actively aligning policy, infrastructure, and incentives to attract AI investment. A moratorium in New York does not make America safer; it makes another country richer in compute capacity.
Moratoriums do not pause demand for computing. They simply determine where that demand is met.
The Strongest Case Against the Thesis
The bear case against FuelCell Energy is not that the power shortage is imaginary. It is that the company is selling a structural story on financials that do not yet reflect it. In the third fiscal quarter ended July 31, 2026, revenue fell 29% year over year to $33 million, pressured by the completion of module deliveries to South Korea and lower generation output. The company posted a net loss of $45.3 million, or 64 cents per share, and a gross loss of $24.5 million, including approximately $17 million in charges tied to inventory write-downs and firm purchase commitments for Phase 0 of its agreement with Fit Energy. Adjusted EBITDA was negative $36.7 million.
The backlog number that has fueled the stock's narrative requires careful parsing. FuelCell Energy reported committed backlog of $1.3 billion as of July 31, 2026, up about 4% from a year earlier, which includes Fit Energy's commitment to purchase systems representing 30 megawatts of generation capacity. The company also added $2.4 billion to a new "awarded capacity backlog" category — but that figure is specifically tied to Fit Energy's option to purchase additional systems representing up to 350 megawatts, not a broad portfolio of signed orders. Combined, the two categories total $3.6 billion. Subsequent to the quarter, the company signed its first Capacity Reservation Agreement for a planned 75-megawatt Texas project, supported by an upfront reservation payment — a reservation, not a construction contract.
Then there is the competitive reality. Bloom Energy is the established leader in fuel-cell data-center power, with signed agreements to power facilities for AEP Ohio, Equinix, and Oracle, and it beat third-quarter expectations while raising full-year guidance in October 2026. FuelCell Energy, by contrast, has no recognized data-center revenue yet. The company is asking investors to underwrite a pivot from a legacy business that is shrinking into a growth business that has not yet delivered a dollar.
This counter-thesis is strong enough that it must be answered with a specific test, not reassurance. The structural-pivot thesis for FuelCell Energy is falsifiable: it requires the company to convert committed and awarded backlog into recognized revenue with positive gross margin on data-center blocks. If quarterly revenue does not approach the $100 million range with positive gross margin by the end of fiscal 2027, or if the Newport Township project or the 75-megawatt Texas reservation are cancelled or delayed beyond 2027, the story reverts to what the current financials show — a cash-burning manufacturer betting on a market a competitor already occupies.
What to Watch and Who Is Exposed
The forward look splits by time horizon. In the short term, FuelCell Energy's stock will trade on narrative momentum and the flow of new capacity reservations. The shares are up about 37% year to date but down roughly 69% from their June 30 peak of $37.88, a swing wide enough to sustain both enthusiasm and disappointment. The stock closed at $11.22 on October 30, well below its 50-day average of $16.07, after falling 7.88% on the day of Few's interview.
Over the medium term, the key signals are contractual rather than rhetorical: conversion of the Texas 75-megawatt reservation into a signed engineering, procurement, and construction contract; revenue recognition beginning on the Fit Energy Phase 0 project, which management has said should start in the fourth fiscal quarter of 2026 and continue through fiscal 2027; and progress toward the 500-megawatt annualized production rate at Torrington.
Over the long term, the question is whether behind-the-meter fuel-cell power becomes a standard component of hyperscale data-center design or remains a niche solution for sites where the grid cannot reach in time. If the structural power shortage persists — and the BloombergNEF and Morgan Stanley shortfall forecasts suggest it will — the addressable market for onsite generation expands regardless of which technology wins. If grid buildout accelerates unexpectedly or AI demand growth slows, the urgency premium embedded in these stocks compresses.
The beneficiaries of the structural thesis extend beyond FuelCell Energy: fuel suppliers (natural gas today, hydrogen blends over time), carbon-capture partners such as ExxonMobil, which is building a Rotterdam carbonate fuel-cell CCS pilot, and electrical-balance-of-plant providers such as Siemens, which is collaborating with FuelCell Energy on optimizing systems for high-density workloads. The exposed parties are utilities and transmission developers whose traditional interconnection model is being bypassed, and jurisdictions whose moratoriums succeed only in exporting investment.
Data as of market close October 30, 2026. FCEL closed at $11.22, down 7.88% on the day.
The real contest in the AI race is not between chip designers; it is between exponential compute demand and linear infrastructure. FuelCell Energy's argument is that the winner will be whoever builds power where the chips are, not wherever the grid happens to be. Moratoriums do not change that math — they only change the address on the invoice.
Explore more exclusive insights at nextfin.ai.
