AI data centers can't wait for the power grid or gas turbines—SOFC is shifting from "backup" to standard equipment.

```

With the surging electricity demand from artificial intelligence data centers and the severe lag in grid expansion, solid oxide fuel cells (SOFCs) are rapidly crossing the commercialization inflection point, rising from a marginal backup option to a standard power source for computing infrastructure.

The core driving force behind this shift is the exhaustion of traditional power supply chains. According to analysis by Yao Yao’s team at GF Securities on June 2, the interconnection queue time for regional grids in the US currently stretches for several years, while the production capacity of major gas turbine manufacturers such as GE Vernova and Siemens Energy has been scheduled out to 2029–2030. For AI data center developers whose construction cycles are only a few months, delays in acquiring power will lead to tens of millions of dollars in capital losses, forcing investors and tech giants to shift capital intensively to distributed independent power sources capable of being delivered within months.

Strong policy interventions are further solidifying this market trend. President Trump’s recent "Electricity Ratepayer Protection Pledge" explicitly requires that new AI data centers must build their own independent power generation resources, prohibiting infrastructure costs from being passed on to ordinary residents. This policy has been signed by seven tech giants, including Oracle, Microsoft, and Google. The establishment of this entry red line, combined with the high tax credit subsidies under the Inflation Reduction Act (IRA), has directly established the rigid demand status of off-grid power generation solutions.

Currently, the capital market is rapidly concentrating on leading companies and core suppliers in the SOFC industry chain. Taking the world's leading SOFC commercializer Bloom Energy as an example, its recent multi-gigawatt framework agreements and strong financial performance indicate that the industry has substantially entered the "1 to 10" phase of scale expansion and performance realization; a value reassessment of the AI power infrastructure supply chain has begun.

Rigid supply of traditional power sources emerges; computing power faces a time gap

The US AI computing race is pushing power demand upward at an unprecedented rate.

According to BlackRock estimates, by 2030 the US will need to add about 148GW of power generation capacity to meet data center demands. However, the pace of grid infrastructure construction is severely lagging: according to PJM Interconnection data, AI infrastructure projects going online in 2025 require on average more than 7 years to reach operational status, with more than 3 years from filing to signing grid interconnection agreements, and another roughly 4 years after approval before full energization. Lead time for large transformers has increased dramatically from about 50 weeks in 2021 to more than 160 weeks by 2026.

The pressure of power shortages is also being transmitted to the residential side. EIA data shows that by 2026 the average residential electricity price in the US will rise to 18.2 cents per kWh, a year-on-year increase of about 5%; according to Bloomberg, in Q1 2026, the average PJM wholesale power price jumped 76% year-on-year to $136.53/MWh, with capacity costs spiking nearly 400%.

Policy catalysts have followed accordingly. In February 2026, Trump's "Electricity Ratepayer Protection Pledge" was instituted, requiring the seven tech giants Amazon, Google, Meta, Microsoft, xAI, Oracle, and OpenAI to sign the pledge to build, introduce, or purchase power resources for new AI data centers at their own expense and bear all infrastructure upgrade costs, prohibiting passing the costs to ordinary residents. This policy directly upgrades "self-generation for self-use" from a commercial option to an entry red line, raising the strategic position of distributed power from an optional supplement to a rigid configuration.

Gas turbine "order overload" becomes SOFC's ticket in

Traditional solutions are not without demand, but rather are severely undersupplied—precisely creating a differentiated substitution window for SOFCs.

As of Q1 2026, GE Vernova’s gas power equipment book orders and reserved capacity have reached 100GW, and are expected to hit at least 110GW by the end of 2026; Siemens Energy’s backlog has risen to 154 billion euros, with a book-to-bill ratio as high as 1.72. The CEO explicitly stated that deliveries have been scheduled to 2029–2030, with very limited supply in 2028. According to Wood Mackenzie estimates, by the end of 2025, global gas turbine orders will have reached 110GW, while global annual manufacturing capacity is only 60–70GW, and in some regions the delivery cycle has reached up to 8 years. The highly overlapping supply chains of the three major companies form a rigid capacity constraint; expansion plans need 3–5 years to take effect, and are bottlenecked by single-crystal blades and other hot-end components supplied by a handful of global vendors.

Small Modular Reactor (SMR) construction cycles require 3–5 years, also unable to meet short-term power needs. Among all practical power options, SOFCs stand out with a 90–120 day delivery cycle (Bloom Energy deployed a MW-class system for Oracle in just 55 days), directly fitting into the "cannot wait" window for AI data centers.

The four technical moats of SOFC: More than just fast

GF Securities believes that speed is only one dimension of SOFC competitiveness; deeper advantages lie in the systematic value brought by the technical architecture itself.

Efficiency advantage: SOFCs have a pure power generation efficiency of 55% to 65%, far exceeding the ~44% of conventional gas turbines; with cogeneration by recovering high-temperature waste heat, the comprehensive energy utilization efficiency can reach 85% to 95%. Bloom Energy’s EnergyServer 5 achieves a self-generation efficiency of 65%, the industry’s highest.

DC architecture dividend: Bloom Energy’s SOFC system directly produces 800V DC power via electrochemical reactions, eliminating the multi-level AC-DC conversions found throughout the power generation and distribution process with traditional gas turbines. For a 1GW AI data center, this architectural change can save about $800 million on UPS systems, $200 million on backup diesel generators, $250–350 million on switching equipment, and $100–150 million on distribution systems—a total capex saving of about $1.35–1.5 billion for auxiliary equipment. These hidden but substantial BOS savings are often overlooked in simple $/kW equipment comparisons yet constitute one of SOFC's strongest economic underpinnings.

Environmental advantages: SOFCs use a non-combustion electrochemical power generation path with zero water consumption, near-zero NOx emissions, and noise levels of just about 65dB (like an air conditioner), suitable for community deployments. Oracle’s Project Jupiter data center, which uses Bloom SOFC, achieved a 92% overall emission reduction and eliminated water use.

Modular scalability: With a 325kW standard module as the basic unit, Bloom Energy can scale up stepwise to hundreds of megawatts or even gigawatt-levels. Power is not interrupted during maintenance and, paired with supercapacitors, can provide second-level load response, supporting over 200,000 charge/discharge cycles and 99.9% availability.

Subsidies plus cost reduction accelerate the economic inflection point

Under the IRA framework, SOFCs as "qualified fuel cell assets" have a base ITC credit of 30%. With domestic manufacturing (+10%) and "energy community" siting (+10%), the maximum credit reaches 50%. Bloom Energy’s latest annual report forecasts a 40% ITC for its projects. The 2025 "Big Beautiful" Act explicitly included fuel cells in Section 48E of the Clean Electricity ITC, with the subsidy window lasting at least through 2032.

On the cost side, current SOFC system costs are about $2,075/kW, with the US Department of Energy targeting costs below $900/kW by 2030. In contrast, gas turbines, driven by short supply, have seen prices soar from around $800/kW in 2021 to about $2,800/kW for 2028–2030 deliveries.

At the power generation cost level, with natural gas prices at $4/MMBtu, a 300MW SOFC yields $0.11/kWh, falling to $0.09/kWh with IRA subsidies; with scale-up to 2.5GW and 50% cost reduction, costs drop further to $0.06/kWh, while gas turbines range from $0.048 to $0.109/kWh. The crossover point for the two cost curves is coming sooner due to continued gas turbine price increases.

Bloom Energy: Leap from GW-level first orders to industry standard

Bloom Energy’s order expansion demonstrates a clear evolution of "icebreaking verification → endorsement by tech giants → establishment as standard solution".

In November 2024, BE signed a 1GW purchase framework agreement with AEP (initial supply 100MW), winning the first utility-grade data center order. In July 2025, BE entered AI hyperscale data center main power supply scenes for the first time, deploying a system for Oracle in just 55 days. In April 2026, Oracle announced its Project Jupiter plan—up to 2.45GW, using 100% Bloom SOFC, fully replacing previous plans for gas turbines and diesel generators, marking the first time SOFC became the "preferred solution" in hyperscale data centers rather than a "backup". In May 2026, BE and AI cloud provider Nebius signed a power rental agreement for up to $2.6 billion and 328MW initially. By Q1 2026, BE’s total orders reached $20 billion, with an order-to-revenue ratio of about 5x, and more than half the backlog was already with other ultra-large cloud service providers, new cloud vendors and host colocation providers besides Oracle, with further diversified customer structure.

On domestic supply chain, several companies have deeply embedded in BE’s system. Sanhuan Group is the world’s leading supplier of separator plates (electrolyte supports) for BE, taking the main position with tape-casting process advantages; Kaizhong Precision has BE certification and has entered mass supply of microchannel heat exchangers; Zhenhua supplies metallic chromium to BE via the supply chain; Jingquanhua and Yilian Technology have entered BE’s supply system for magnetic components and electrical connection units respectively, both likely to directly benefit from BE’s order ramp-up.

Domestic enterprises with independent system capability are also actively deploying. Weichai Power in November 2025 obtained a license from Ceres Power UK to manufacture metallic-supported SOFC technology, aimed at AI data centers; in 2025, data center power product sales were about 1,400 units, a 259% year-on-year increase. Yishitong is pushing full industry chain R&D; its 120kW demonstration project is due to launch in Q2 2026 and its 1GW SOC project is under construction. Foran Energy is advancing 50–300kW SOFC system demonstration applications; Binglun Environment is involved in a China Mobile Guizhou data center pilot project.

Risk warning and disclaimerThe market has risks. Investment must be prudent. This article does not constitute individual investment advice nor does it take into account special investment objectives, financial status, or needs of individual users. Users should consider if any opinions, views, or conclusions in this article are suitable for their specific situation. Investing based on this is at your own risk. ```