Powering data centers: which startup is ahead?

Which companies count as data-center power startups?

The serious private data-center power startup field currently contains seven companies, plus Fervo Energy as a public-market benchmark. Their products range from commercial generators to nuclear reactors that remain several years from supplying customers.

We include startups developing dedicated, firm or dispatchable electricity for large data centers. That covers on-site generators, advanced geothermal, dispatchable solar and small nuclear reactors. We exclude batteries sold mainly for short-duration backup, grid-management software, cooling systems, utilities and established equipment groups such as Caterpillar, Cummins, GE Vernova and Wärtsilä.

Funding totals need a careful read. Some companies announce equity only, while others combine equity, debt and government support. Kairos Power has disclosed a large Department of Energy agreement without publishing a complete cumulative private-funding figure. Last Energy has also announced its latest round without providing a consistently verifiable lifetime total.

Company	What it is building	Known cumulative capital	Current position
Mainspring Energy	Modular, fuel-flexible linear generators for on-site power	More than $800M after a $258M Series F	Commercial product and current private leader
Sage Geosystems	Pressure-geothermal generation and underground energy storage	About $159M, including a $97M Series B	Commercial storage operating; first generation plant due in 2027
Kairos Power	Fluoride salt-cooled advanced nuclear reactors	Private total undisclosed; DOE support agreement worth up to $303M	First commercial-scale reactor under construction
Exowatt	Solar-thermal generation with integrated long-duration heat storage	About $140M	First pilot expected by the end of 2026
Valar Atomics	High-temperature gas microreactors for industrial and AI campuses	At least $580M reported across recent financings, including debt	Reactor demonstration completed; commercial licensing unresolved
Aalo Atomics	Factory-produced modular nuclear plants for data centers	More than $136M	Test reactor reached criticality; factory expansion planned
Last Energy	Factory-built 20 MW pressurized-water microreactors	More than $100M in its latest Series C; total undisclosed	Large announced project pipeline, with no operating reactor
Fervo Energy	Enhanced geothermal power using horizontal drilling	More than $1B before its IPO, followed by about $2.2B in IPO proceeds	Public benchmark, excluded from the private-startup ranking

Is Mainspring Energy really ahead in data-center power today?

Mainspring Energy is currently the most advanced private data-center power startup because it already sells working hardware while its closest challengers are still building their first commercial generation projects.

Mainspring began commercial shipments in 2020. The company now reports hundreds of megawatts across field operations and advanced development for Fortune 500 companies, utilities and data-center developers. That figure combines operating equipment with future projects, so it should not be read as hundreds of megawatts already powering data centers. Even so, no other private company in the comparison has the same record of repeated commercial hardware deliveries.

The strongest new proof is a 48 MW project ordered by the Utah Municipal Power Agency. The facility is scheduled to begin operating in 2027 and moves Mainspring beyond scattered 250 kW installations into utility-scale generation. One 48 MW project is also larger than the entire operating output of most emerging power technologies in this field.

Sage is behind on current generation but has a credible route toward closing the gap. Meta selected its geothermal technology for a project intended to reach 150 MW, and Ormat is helping develop Sage’s first commercial power plant in Nevada. Kairos has an even larger long-term opportunity with Google, although its first electricity remains scheduled for the end of the decade.

The gap between first and second place is meaningful, not huge. Mainspring leads private startups in what customers can buy now. Sage could become the stronger clean baseload supplier if its first generation plant works as planned.

If you want more recent data on this point, please see our latest data center market report.

Which data-center power startup has the strongest commercial traction?

Mainspring has the strongest current commercial traction, although the company still discloses far less data-center detail than it should.

Real commercial adoption appears in several forms. Lineage ordered 33 additional linear generators for five Texas facilities after using Mainspring systems at sites in California and the Northeast. A separate CalBio deployment has grown to 5.3 MW across five locations. These are industrial rather than AI data-center installations, but they show customers expanding the product after earlier deployments instead of stopping after one demonstration.

The data-center evidence remains thinner. Industry reporting said Mainspring expected its first full-scale greenfield AI data-center project to come online in 2025 and eventually expand into the hundreds of megawatts. The company has since avoided naming the customer, operating capacity or location publicly. Until those details emerge, that project cannot be treated like an independently verified hyperscale deployment.

Sage has less revenue-producing hardware but stronger disclosed data-center customer evidence. Meta has publicly confirmed the relationship, while Sage’s own project roadmap places the 150 MW development around 2030. Sage has also commissioned a 3 MW underground energy-storage installation, giving the company some operating field experience before it attempts full commercial power generation.

Kairos has achieved something different: contractual credibility before commercial operation. Its 50 MW Hermes 2 reactor has a binding power-purchase agreement with TVA and will support Google’s data centers in Tennessee and Alabama. That agreement carries more weight than a reservation or memorandum of understanding, but revenue-producing electricity is still years away.

Fervo shows what mature traction looks like in this category. Banks committed $421 million of non-recourse project financing to Cape Station, meaning lenders evaluated the project largely against its own contracts and expected cash flows. None of the remaining private startups has reached that level of project bankability.

Which startup can power a new data center fastest?

Mainspring currently offers the fastest credible route from an available site to firm on-site data-center power.

Each Mainspring unit produces 250 kW and uses the same basic design across different project sizes. A developer can begin with a smaller installation, add modules as buildings open and connect to the wider grid later. The generators can also operate independently from the grid, directly addressing interconnection queues and delayed substation upgrades.

A 100 MW campus would need roughly 400 modules. That is a big order, but it involves factory production and repeated installation rather than drilling a geothermal field or licensing a nuclear reactor. Customers can also phase the equipment alongside the data-center build instead of waiting for the entire 100 MW block.

Exowatt aims to offer a similarly integrated route through ExoRise, which combines suitable land, modular data-center shells and solar-thermal power. Its first pilot is expected by the end of 2026, so today the proposition remains a development platform rather than a repeatedly installed power product.

Sage’s first commercial generation facility is targeted for 2027. Kairos plans first delivery by 2030, while Aalo, Valar and Last Energy have yet to provide a commercial schedule supported by the same combination of licensing progress, construction and a binding power buyer.

Mainspring’s speed advantage comes with a fuel constraint. Most near-term projects will probably run on natural gas because renewable hydrogen and biogas are unavailable at the quantities required by a hyperscale campus. Developers still need firm gas capacity, local permits and community acceptance. It is fast, not instant.

Which startup has won the best data-center customers and contracts?

Kairos has secured the strongest contract structure, while Sage has won the most valuable customer relative to its current size.

Kairos has linked three parties that normally sit separately in an advanced-energy announcement. Kairos builds the reactor, TVA buys up to 50 MW through a binding PPA, and the electricity supports Google data centers. Construction started on Hermes 2 in April 2026, giving the agreement a named site and a physical project rather than only a future technology promise.

Sage’s Meta partnership is commercially important for a company with roughly $159 million of funding. The planned project is several times larger than Sage’s present operating storage installation and gives the startup access to one of the world’s biggest data-center buyers. Ormat’s involvement adds an experienced geothermal developer that can handle parts of construction and plant operation Sage has never performed at this scale.

Mainspring’s customer base is broader, with names such as Lineage, NextEra Energy Resources, utilities and the U.S. Air Force. The $150 million NextEra agreement created a channel for equipment purchases and customer financing, while Schneider Electric offers Mainspring as part of an integrated microgrid solution. Those relationships make the company easier to buy from, finance and maintain.

Mainspring loses points because its most important data-center buyers remain undisclosed. A named utility project is stronger evidence than an unnamed hyperscale opportunity, even when management says the opportunity is large.

Valar’s work with Nvidia has generated the most attention lately. The companies demonstrated a microreactor system powering Nvidia hardware and discussed a 30 MW closed-loop AI facility in Utah. Public descriptions still frame the project as a collaboration to explore or develop the concept, leaving the commercial commitment less certain than the Kairos-TVA agreement.

If you want more recent data on this point, please see our latest data center market report.

Which startup has the largest believable data-center power pipeline?

Mainspring has the largest near-term private pipeline we can take seriously, but only a fraction of it is publicly visible at project level.

Latitude Media reported that Mainspring had assembled a $1 billion data-center pipeline by July 2026. The company says it screens speculative proposals and focuses on developers with credible sites, gas access and community plans. That makes the number more useful, though a pipeline still includes projects that may never become firm orders.

Sage’s 150 MW Meta project is smaller but easier to understand. We know the customer, technology and broad delivery window. The project deserves a higher confidence weighting per megawatt than capacity mentioned without a site or buyer.

Kairos and Google have a broader agreement to develop up to 500 MW across multiple reactors. The first 50 MW is much more believable than the remaining 450 MW because Hermes 2 has entered construction and sits under a binding TVA PPA. Later reactors depend on the first plant proving its technology, construction methods and economics.

Exowatt reports more than 90 GWh of signed demand. The figure sounds large but measures energy over time rather than generating capacity. Spread evenly across one year, 90 GWh equals an average load of about 10.3 MW. The true capacity could differ substantially depending on contract length and delivery profile, neither of which Exowatt has disclosed.

Last Energy has proposed 30 reactors for a 600 MW Texas campus serving data-center customers. Its 20 MW modular design makes the arithmetic straightforward, although the project remains at the planning stage. Until licensing, financing and customer commitments become more visible, the 600 MW figure belongs in a development pipeline rather than a contracted-capacity comparison.

Fervo shows how wide the gap is between a pipeline and a mature order book. The now-public company reports 658 MW of contracted PPAs alongside a separate Google framework covering up to 3 GW. We count the 658 MW far more heavily because executed PPAs can support financing and construction.

Which data-center power startup is moving fastest lately?

Valar has produced the sharpest recent technical acceleration. Mainspring has built the stronger commercial momentum.

Valar moved from zero-power criticality in late 2025 to a live 2026 demonstration in which its Ward 250 system produced roughly 100 kW of thermal energy and powered Nvidia computing hardware through a thermoelectric generator. The startup also raised a reported $450 million financing package after its earlier $130 million Series A. Few reactor companies have compressed that much funding, hardware progress and public attention into such a short period.

The scale difference is enormous. A 100 kW thermal demonstration produces far less usable electricity than a 30 MW data center would consume. Commercial operation will also require a licensing path that goes beyond the Department of Energy research program used for the test.

Mainspring’s recent progress has been less dramatic but more relevant to buyers. The company signed the 48 MW UMPA project, appointed former Cummins chairman and CEO Tom Linebarger as chief executive and continued preparing its large Pennsylvania manufacturing facility. Those moves look like a shift from founder-led commercialization toward industrial execution.

Aalo also reached criticality in 2026 and announced plans for a one-million-square-foot manufacturing site. Criticality validates important parts of reactor physics and operating procedures. Commercial electricity still requires heat removal, power conversion, licensing and reliable plant operation.

Sage’s $97 million Series B may prove more important than a louder technical demonstration. Ormat co-led the round and committed to Sage’s first commercial generation facility, joining capital and project execution in the same transaction. That is how a laboratory technology starts becoming a power business.

If you want more recent data on this point, please see our latest data center market report.

How mature are these data-center power products?

Mainspring is the only private company in the group with a repeatedly shipped generation product. The nuclear startups remain much earlier than their recent demonstrations suggest.

Sage has crossed from laboratory work into field operation through its underground storage project, although pressure-geothermal electricity generation is still being tested. Exowatt has manufactured early P3 systems and gathered customer demand but has yet to operate its first ExoRise pilot.

Kairos is the most mature nuclear developer in the group because Hermes 1 is under construction and Hermes 2 has received an NRC construction permit. Aalo and Valar have achieved criticality faster, yet those milestones came through research-oriented test programs rather than a licensed commercial electricity plant.

The maturity gap inside nuclear is easy to miss. Criticality proves that a sustained fission reaction can occur. A data-center customer needs a complete facility that converts heat into electricity every hour, passes commercial regulation, receives fuel reliably and can be serviced for decades.

Company	Product stage now	Best proof available	Remaining leap
Mainspring	Commercial shipments and expanding installations	Repeat customers, utility project and multi-sector field operations	Prove a named hyperscale data-center deployment
Sage	Commercial storage operating; generation in testing	3 MW storage project, Meta relationship and Ormat-backed plant	Operate the first commercial pressure-geothermal generator
Kairos	Commercial-scale demonstration reactor under construction	NRC construction permit, TVA PPA and Google agreement	Finish construction and deliver electricity by 2030
Exowatt	Pilot-stage power platform	Early P3 systems and 90+ GWh stated demand	Demonstrate field output, reliability and manufacturing economics
Valar	Powered reactor demonstration	Criticality and live Nvidia hardware test	Secure commercial licensing and build a customer-scale plant
Aalo	Critical test reactor operating	Commercial-scale criticality milestone	Add power conversion, licensing and repeated manufacturing
Last Energy	Development and pre-deployment	20 MW design, funding and proposed Texas campus	License, finance and construct the first operating unit

Which startup gives data centers the cleanest 24/7 power?

Sage currently has the strongest private-startup proposition for future round-the-clock carbon-free power, although Kairos could eventually offer greater geographic flexibility.

Pressure geothermal can provide continuous electricity without depending on weather or daily battery charging. Sage also claims a relatively small surface footprint and access to hot dry rock beyond traditional geothermal regions. Its first commercial generation project will show whether those advantages survive outside testing.

Kairos, Aalo, Valar and Last Energy could also supply high-density carbon-free electricity around the clock. Nuclear fuel requires very little land relative to its energy output, which makes the technology attractive for large campuses. Timing is the problem. Kairos has the most credible delivery plan, and even that project is targeting 2030.

Exowatt combines solar collection with thermal storage designed to deliver power during the night. The company describes the output as firm and dispatchable, but its first pilot has yet to show how the system performs across changing weather, seasons and repeated daily cycling.

Mainspring can operate around the clock immediately, though most projects will run on natural gas. The generator produces very low local nitrogen-oxide emissions and can switch among natural gas, propane, biogas and hydrogen. Its carbon footprint depends far more on the fuel supplied than on the generator’s theoretical flexibility.

Buyers currently have to choose between deployability and carbon intensity. Mainspring can solve a power shortage now. Sage and the nuclear startups promise cleaner long-term supply, with more development risk attached.

Which startup offers data centers the best economics?

Mainspring offers the clearest economic value today because it can bring expensive computing capacity online earlier, even though we cannot prove that its electricity has the lowest lifetime cost.

Data-center power economics go beyond cents per kilowatt-hour. A campus waiting three years for grid capacity may lose far more through delayed rental revenue or unused GPUs than it spends on a higher-cost on-site generation system. Mainspring can win projects through avoided delay without beating every utility tariff.

Its modular structure also reduces the need to build a complete power plant before the first data hall opens. Capacity can follow construction in 250 kW increments, limiting the amount of generation sitting unused during early phases.

The missing data is important. Mainspring does not publish a standard data-center electricity price, verified maintenance cost or comparable lifetime cost for large projects. Natural-gas prices, financing, operating hours and permitting conditions will change the result from one site to another.

Sage could ultimately produce cheaper carbon-free baseload power by using existing drilling, oilfield and geothermal supply chains. Today there is too little commercial generation data to calculate a dependable levelized cost.

Exowatt has discussed ambitious cost targets for solar-thermal power. Those targets depend on manufacturing scale the company has yet to demonstrate. The first pilot will tell us more than another theoretical estimate.

Nuclear economics are even harder to judge. Aalo promotes a long-term target of three cents per kWh, while Last Energy markets standardized 20 MW plants with private financing. Neither figure has been tested through an operating commercial fleet.

We rank Mainspring first on present customer economics, especially where time-to-power is the dominant cost. Sage has the strongest chance of beating it on long-term clean-power economics after commercial generation begins.

If you want more recent data on this point, please see our latest data center market report.

Which startup can actually deliver hundreds of megawatts?

Mainspring has the strongest manufacturing plan, but its announced factory would still struggle to satisfy multiple gigawatt-scale data-center orders at once.

The planned Pennsylvania facility will have capacity for as many as 1,000 linear generators per year. At 250 kW each, that equals 250 MW of annual nameplate production. The factory is supported by an $87 million Department of Energy grant and is expected to employ more than 600 people once operating.

A 250 MW annual rate would represent genuine industrial scale for a startup. It would also cover only one quarter of a 1 GW campus. Mainspring will need more factories, manufacturing partners or substantial production expansion if several hyperscale customers order equipment simultaneously.

Sage uses a project-development model rather than a product factory. Its ability to scale will depend on drilling crews, project finance, suitable geology, permits and partners such as Ormat. That path can support large individual projects but may grow less uniformly across regions.

Kairos is building reactor equipment modules at its Albuquerque manufacturing campus before shipping them to Oak Ridge for assembly. Hermes 1 and Hermes 2 should show how the modular construction process behaves under real nuclear-quality requirements. Repetition remains a future goal until the first power-producing plant is finished.

Aalo’s proposed one-million-square-foot factory is the boldest manufacturing plan in the nuclear group. The company may eventually gain an advantage by building factories and reactor designs together. It also carries a risk: investing heavily in production before the final commercial design and regulatory pathway stabilize can force expensive changes later.

Exowatt faces a similar volume question. Modular solar-thermal hardware could scale quickly once standardized, but the company has not disclosed a demonstrated annual output rate. Today, Mainspring is the only private contender whose production plan can be translated into a credible annual megawatt figure.

What can Mainspring do that competitors cannot easily copy?

Mainspring’s advantage comes from its linear-generator design, years of field learning and an emerging sales ecosystem rather than simply placing a generator beside a data center.

The system converts fuel into electricity through a low-temperature, flameless reaction that moves magnets through copper coils. It avoids lubricating oil and many of the complex moving components found in conventional engines. Customers can also switch gaseous fuels without replacing the equipment.

The larger moat may come from operational history. Mainspring has shipped equipment since 2020, giving it years of information on maintenance, component wear, fuel variation and customer load profiles. A rival can design a competing machine, but reproducing several years of operating data takes time.

NextEra helps finance and deploy customer projects. Schneider Electric integrates Mainspring systems into microgrids. Resellers extend the company into additional industries and regions. These relationships reduce the friction of buying from a relatively young manufacturer.

Established power-equipment companies remain a major threat. Caterpillar, Cummins and other incumbents have huge service networks, trusted balance sheets and factories built for high-volume production. Mainspring must offer enough improvement in emissions, maintenance and fuel flexibility to justify choosing a younger supplier.

Sage’s moat is different. The company combines subsurface models, well-design knowledge and intellectual property around pressure geothermal. Successful operation with Meta and Ormat would deepen that advantage because each project generates geological and operating data.

Kairos could eventually develop the strongest regulatory moat. A licensed reactor design, qualified fuel supply and experienced construction workforce would be difficult to reproduce. That protection becomes commercially valuable only after electricity reaches customers.

How much can we trust the data behind this startup ranking?

Mainspring leads on operating evidence but still leaves major disclosure gaps. Kairos provides unusually strong contract and regulatory evidence for a pre-revenue company.

We place the greatest weight on equipment operating at customer sites, repeat orders, binding PPAs, construction permits, project financing and named counterparties. Company pipelines, capacity reservations and long-term cost targets receive less weight.

The distinction changes several rankings. Exowatt’s 90 GWh backlog sounds more advanced than Aalo’s test reactor, but the backlog lacks a delivery schedule. Aalo’s criticality milestone is technically concrete, though it says little about commercial electricity costs. The two facts answer different questions and should not be blended into one generic measure of progress.

Mainspring’s broad claim of hundreds of megawatts combines field operations with advanced development. We rely more heavily on specific disclosed projects such as UMPA’s 48 MW facility and Lineage’s 33-unit expansion.

Valar’s recent demonstration deserves credit for producing heat and powering computing equipment. Its proposed Nvidia facility carries less weight until the companies disclose a commercial contract, licensing plan, construction schedule and committed capital.

Private funding totals are uneven too. Mainspring’s published total includes more than $800 million of financing. Valar’s reported $450 million package mixes $340 million of equity with $110 million of debt. Kairos has received large government support alongside private capital that remains undisclosed.

Company	Evidence quality	Strongest public evidence	Biggest missing fact
Mainspring	Medium-high	Commercial shipments, repeat customers, 48 MW utility project and funded factory	Named operating data-center capacity and revenue
Sage	Medium	Operating storage, Meta relationship, Ormat partnership and funded 2027 plant	Commercial geothermal generation performance
Kairos	Medium-high	NRC construction permit, active construction and binding TVA PPA	Actual construction cost and operating output
Exowatt	Medium-low	Funding, early hardware and signed-demand claim	Pilot performance, contract terms and factory output
Valar	Medium	Criticality, live powered demonstration and major financing	Commercial license, binding customer order and plant schedule
Aalo	Medium	Criticality and manufacturing investment	Power-producing reactor performance and committed customers
Last Energy	Medium-low	Financing and announced project pipeline	Licensing progress, firm contracts and constructed hardware

Which startups are actually ahead in powering data centers?

Mainspring Energy is the startup ahead overall today, with Sage Geosystems as the closest challenger and Kairos Power leading the longer-term nuclear race.

Mainspring wins because it has crossed the hardest early threshold in energy hardware: customers can already buy and operate the product. The company has repeat installations, a 48 MW utility project, a credible manufacturing expansion and a direct answer to the data-center industry’s immediate problem of waiting years for grid power.

Its lead comes with two weaknesses. Mainspring has not publicly broken out operating data-center megawatts, and most near-term projects will depend on fossil natural gas. The company is ahead commercially without owning the cleanest long-term position.

Sage ranks second because the combination of Meta, Ormat, commercial storage experience and a funded generation project gives it the clearest path toward challenging Mainspring. A successful 2027 plant would transform Sage from a promising geothermal developer into a real power producer.

Kairos ranks third. Its first plant has a construction permit, active groundwork, a utility buyer and a hyperscale end customer. No other private nuclear startup has assembled those four pieces. Its 2030 delivery schedule keeps it below companies capable of affecting current data-center construction.

Exowatt comes fourth. A pilot expected this year and a substantial stated demand backlog put it closer to customers than most reactor startups. We still need proof that its solar-thermal system can operate reliably and be manufactured at attractive cost.

Valar ranks fifth after moving extraordinarily quickly through fundraising, criticality and a live Nvidia demonstration. It could climb fast, although the current evidence says more about engineering speed than commercial power delivery.

Aalo follows in sixth place. Its criticality milestone and factory plans are serious, but Kairos remains further along in regulation, construction and customer contracting. Last Energy ranks seventh because its proposed pipeline is large while visible project execution remains limited.

Fervo would rank above every company here on overall commercial maturity. Its transition to the public market is precisely why Mainspring now holds the startup title.

The ranking could change under three clear conditions. Mainspring would extend its lead by disclosing a large operating AI data center. Sage could take first place after successfully commissioning commercial pressure-geothermal generation and moving Meta’s 150 MW project into construction. Kairos could eventually pass both if Hermes 2 reaches operation on schedule and becomes a repeatable template for the rest of Google’s reactor fleet.

Rank	Startup	Why it is ahead
1	Mainspring Energy	Only private contender with a repeatedly shipped product, repeat customers, a 48 MW utility project and a credible route to fast on-site data-center power
2	Sage Geosystems	Strongest clean-power challenger, backed by Meta, Ormat and an operating storage project, with first commercial generation planned for 2027
3	Kairos Power	Best nuclear contract and regulatory position through TVA, Google, an NRC permit and active Hermes 2 construction
4	Exowatt	Near-term pilot and meaningful stated demand, balanced against unproven field economics and manufacturing scale
5	Valar Atomics	Fastest recent technical momentum and deep funding, with commercial licensing and customer commitments still unclear
6	Aalo Atomics	Reached criticality and is investing aggressively in production, but remains behind Kairos in construction and customer evidence
7	Last Energy	Large proposed data-center pipeline and familiar reactor architecture, with limited visible progress toward an operating commercial unit

If you want more recent data on this point, please see our latest data center market report.

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.