Bloom Energy Gains as Data Center Pushback Forces On-Site Power

Local backlash is changing how AI gets powered

One of the companies benefiting most from the political fight over AI infrastructure doesn't build data centers. It builds the power plants that let developers avoid waiting on, or arguing with, the local utility. Bloom Energy makes solid oxide fuel cells that generate electricity on site. Its business has grown sharply this year, and across the coverage the cause is the same: more communities are refusing to let new data centers draw on a shared grid that is already strained.

Bloom's numbers show how much has changed. Second-quarter 2026 revenue was about $1.06 billion, up 166% from a year earlier. The company posted a 17.1% operating margin and $175 million in free cash flow, then raised its full-year revenue guidance to $3.9–4.2 billion.23 Product revenue alone was roughly $935 million, more than three times the year-earlier figure.21 Bloom swung from a $3.5 million operating loss a year ago to a $182 million operating profit.25 Brookfield has expanded its financing framework for Bloom-powered AI projects fivefold, to $25 billion.28 Oracle has agreed to buy up to 2.8 gigawatts of Bloom systems.27

Wall Street has rewarded the run, and also made it fragile. The stock gained about 235% in the twelve months through early October.25 Some reporting now puts it at around 56 times expected 2027 earnings.21

The backlash, measured

Opposition to data centers is now a measurable obstacle for developers. Bloomberg, citing Data Center Watch, reports that local resistance blocked or delayed at least 120 projects worth almost $200 billion in the first half of 2026. The issue has become a theme in the midterm campaigns.3 A Gallup poll found that 71% of Americans oppose an AI data center in their own community. Only 53% said the same about a nuclear plant.5 Separate counts differ on how much investment has been hit, but they point the same way. STL Partners estimates about $77 billion of U.S. data center investment has been delayed or cancelled.1 Another tally counts at least 75 projects worth about $130 billion halted in the first quarter alone.5

Electricity bills are the main complaint. In the PJM region, which serves 65 million people, capacity auction prices jumped from $28.92 to $329.17 per megawatt-day. The grid operator attributed 63% of that rise to data center demand.4 State governments have started acting on that anger. New York paused environmental approvals for data centers above 50 megawatts. Texas froze new data center projects in its interconnection queue pending audits. Pennsylvania now requires developers to supply their own power and pay their own electricity costs.5

The resistance is also spreading abroad. More than 70 European projects were rejected or restricted between January and April, more than in all of 2025. Scotland has paused approvals for hyperscale sites, and Denmark has moved data centers toward the back of the grid queue.1

Why pushback helps Bloom

The reporting splits on what the backlash means for the overall buildout. Goldman Sachs says the pipeline of planned capacity is large enough to absorb project losses. The bank raised its end-2026 U.S. capacity forecast to 64 gigawatts while trimming its 2027 view. It argues that power and land, not opposition, are the real constraints.28 Sightline Climate is less optimistic. It expects 30% to 50% of large data centers scheduled to open this year to be delayed or cancelled, with only 5 of 16 announced gigawatts actually under construction.6

Both views lead to the same conclusion. Goldman says developers are going "where the power is"8, and more and more, that means power they build themselves. One developer told TechRepublic that a project without its own generation plan is usually a waste of time. Grid capacity can't keep up, so on-site generation paid for by the developer has become standard.10 Pennsylvania's self-supply mandate writes that expectation into law.5

Bloom fits that need. Its fuel cells sit on site and avoid substations, transmission lines and interconnection queues.24 Those queues can run three to seven years in major utility territories. High-power transformers now take three to five years to arrive, and switchgear is reportedly sold out through 2028.6 Oracle explained its expanded deal in terms of speed, not cost.2127 Bloom CEO KR Sridhar says every major U.S. hyperscaler has now approved Bloom's systems for AI facilities.24 One analysis summed up the pitch: in AI, the costliest thing isn't electricity but waiting for it.29

There's a political benefit too, even if Bloom's materials don't stress it. A data center that makes its own power doesn't add to the shared grid upgrades that drive up household bills. That is exactly the complaint fueling local opposition.24 In our view, Bloom's growth comes less from AI demand in general than from regulators and voters deciding that data centers should pay for, and supply, their own power.

Nuclear is the long-term answer, and it's slow

The big nuclear deals of the past month show the timing gap Bloom is filling. Google signed a 20-year agreement with Constellation for 890 megawatts of new output from upgrades at 11 reactors. Constellation will spend more than $4.3 billion on the work. A separate 15-year contract covers 2,700 megawatts from Constellation's existing fleet.1115 The first upgraded unit isn't expected to deliver power until 2028, and work continues into the early 2030s.12 Days earlier, Amazon signed for 690 megawatts tied to Calvert Cliffs, with that plant's upgrade coming online between 2030 and 2032.20 Oracle wants to buy 10–20% of the output of Wisconsin's Point Beach plant for its Port Washington campus. It said the power is valuable because it's "available now," even if it's expensive.13

These deals show that hyperscalers want firm, clean, around-the-clock power. They also show how far away it is. About 8.5 gigawatts of existing U.S. reactor output is now contracted to big tech companies. Only one new reactor built for data centers, the 50-megawatt Kairos Hermes 2, is under construction.20 Even small modular reactors are seen as competitors mainly in the 2030s, which limits their effect on projects scheduled for 2026–2030.29 Google's energy chief framed the Constellation deal around speed, and noted that data centers are "under a lot of criticism."12 Nuclear helps on public perception and on long-term supply. Through the end of the decade, on-site gas is what can actually be built quickly.

Where the story is weaker

Natural gas is Bloom's biggest risk. Its fuel cells need a steady supply. The 2.45-gigawatt Project Jupiter in New Mexico, Bloom's largest single deployment, depends on an Energy Transfer pipeline that has slipped from August 2026 to February 2027. Oracle issued a force majeure notice, project debt has traded below 90 cents on the dollar, and Bloom shares fell 8% when the news broke.26 Bloom says Oracle remains committed. Oracle says the project is still on schedule.21 Bloom also says its equipment can be moved to other projects and that its 2026 guidance doesn't depend on any single site.26

To us, Jupiter is more than a one-off delay. Building your own power takes a data center off the grid, but not off pipelines, permits and state regulators. The opposition can move upstream to the gas supply.

Profitability is less certain than revenue growth. One account puts Bloom's latest product gross margin at 36.5%, essentially unchanged from 2024.21 Another cites 37.2%.25 Either way, Bloom is selling far more systems without yet keeping a much larger share of each sale. Roughly all of Bloom's first-half operating cash flow matched inflows from customer deposits and deferred revenue, which means customers are paying ahead of delivery.29

Bloom's newest pitch is that its fuel cells natively produce 800-volt DC power. It says that could cut $3.6 billion, or 27%, from the non-chip construction cost of a 1-gigawatt AI campus.22 That figure comes from Bloom's own model, and no customer has been named for the full design. Nvidia's list of 800-volt partners included Eaton and GE Vernova but not Bloom.21

The bottom line

On the main question, the coverage agrees: the AI buildout is running into a grid it can't wait for, and into neighbors who don't want to pay for it. Nuclear is where hyperscalers expect to get their long-term clean power, but deliveries are measured in years. Local and state pushback isn't stopping AI construction. It is forcing it onto self-supplied power. That makes Bloom one of the clearest near-term winners, as long as it can secure gas supply and deliver on a valuation that leaves little room for delays.