Energy · Deep dive

Exowatt

A Miami-born, Atomic-incubated energy startup selling the P3 — a 40-foot containerized module that concentrates sunlight through Fresnel lenses into a thermal battery and dispatches electricity through a Stirling engine — pitched as near-free, 24-hour dispatchable solar for AI data centers in the sun-soaked US Southwest.

emerging

The question that decides it: Exowatt's entire economic claim rests on a Fresnel-lens-plus-heat-battery-plus-Stirling-engine module hitting an unsubsidized levelized cost below $0.04/kWh today and $0.01-0.02/kWh at roughly a million units a year of production — cheaper than commodity solar-plus-storage. But Lazard already put unsubsidized utility-scale PV as low as $0.024/kWh in 2023, lithium-ion pack prices keep falling on a global supply chain no startup can match, and every prior company built on the same ingredients — Stirling Energy Systems (bankrupt 2011), Infinia (bankrupt 2013), modular-CSP pioneer Ausra (shuttered 2014) — died when PV costs collapsed. So the question is mechanical and falsifiable: can a proprietary, moving-parts thermal architecture with a ~25 kW-per-container output and 1,000-plus-acre footprints per 50 MW of 24-hour load actually out-cost the PV+BESS commodity curve at production scale before Exowatt's capital runs out — or does the January 2026 pivot into ExoRise powered-land development concede that the module alone was never going to win, and the real product is desert land with an interconnection story?

Snapshot

Exowatt is a Miami startup, incubated inside Jack Abraham’s venture studio Atomic in 2023, that sells the P3: a bright-orange, 40-foot-container-sized module that concentrates sunlight through Fresnel lenses into a proprietary thermal battery and converts the stored heat to electricity on demand through a Stirling engine. It claims the P3 can deliver up to 24 hours of dispatchable power at just under $0.04 per kWh today, falling to $0.01-0.02 at mass production — aimed squarely at AI data centers stranded in interconnection queues. It has raised roughly $140M (Sam Altman and Leonardo DiCaprio in the seed; Felicis leading the Series A) and claims a 90+ GWh demand backlog. What it does not yet have, three years and $140M in, is a publicly verified operating commercial deployment: first pilots began mid-2025, and its first ExoRise powered-land site is slated to be operational by end of 2026. The gap between the claims and the commissioned steel is the whole story.

Founding story

The idea came from the investor, not the engineer. Abraham — Atomic’s CEO, a co-founder of Hims & Hers — saw a heliostat concentrated-solar plant on a trip to Israel and developed what he calls a thesis for making CSP viable at modular scale. He called Hannan Happi, whom he’d met at an Atomic event, and invited him to Miami to build it in early 2023.

Happi’s story deserves the unvarnished version. A TU Munich-trained mechanical engineer with a Stanford GSB degree, he spent time at GE, Siemens, Accenture and about a year as an operations analyst at Tesla — a thin direct-energy résumé, as Latitude Media pointed out. His main act was Volansi, a VTOL cargo-drone company he co-founded in 2015 and ran as CEO; it raised roughly $75M, lost its venture backing, and ceased operations in 2022, with Sierra Nevada Corp buying the assets that October in a deal Happi did not join. Through Volansi and even Exowatt’s September 2024 launch party he went by Hannan Parvizian; more recent coverage and his LinkedIn use Hannan Happi. (A rumor sometimes attaches him to Rivian; the record shows Volansi, Tesla, Siemens and GE — no Rivian.) He says the team iterated through more than 50 designs in 2023 before settling on the P3’s lens-battery-engine combination. Whatever one thinks of the architecture, the founder-market fit here is a hardware operator plus a studio with zero prior energy investments — a fact the energy trade press noticed immediately.

How it works

Each P3 is a factory-built, bolted-steel module on a 40-foot-container footprint. Sunlight enters through Exowatt’s proprietary Fresnel lenses on top and is concentrated onto long-duration heat-battery cells made of what the company calls proprietary silicon-based composite materials — media descriptions have ranged from fumed silica to a bed of domestically sourced sand and slag heated toward 1,000°C. Five cells make a pack; two packs feed one Stirling engine, a centuries-old external-heat engine that circulates a working gas to drive a generator. Each module dispatches about 25 kW of electricity at 480 V, for up to 24 hours per day of dispatch, and the cells can also be charged by built-in resistive heaters running on cheap grid power — a hedge that quietly makes the P3 part electro-thermal battery, not purely solar. Configuration is modular: more engines for more power, more packs for more duration.

The physics constraint is land. By Exowatt’s own online calculator (September 2024), serving a 50 MW data-center load around the clock in Arizona would take up to 1,123 acres of modules — roughly four times the footprint of an equivalent PV farm — and concentrated solar only works economically in narrow high-direct-sunlight desert bands, which is why everything Exowatt plans sits in West Texas, New Mexico, Arizona and Nevada.

Product and business overview

There are now two named businesses. The Exowatt P3 is the hardware: an integrated capture-store-dispatch module sold (or reserved) directly to data-center operators, energy developers and industrial customers, with US-sourced components positioned to capture Inflation Reduction Act credits and dodge tariff risk. ExoRise, launched January 21, 2026, is a powered-land development arm: Exowatt acquires desert acreage in the Southwest — it markets the region as “Frontier Land” — and delivers turnkey, islanded, behind-the-meter campuses (land, permits, P3 power) for hyperscalers who cannot wait five years in an ERCOT or PJM interconnection queue. ExoRise is a substantial strategic shift, as Latitude Media noted: a hardware vendor becoming a site developer eighteen months after launch is also a signal about how the hardware was selling on its own.

Business model and pricing

The original model was direct unit sales, with an online reservation queue and a $10,000 reservation fee. Real price points surfaced via Exowatt’s own calculator in September 2024: a 50 MW, 24-hour-dispatch Arizona configuration priced up to $420M in capex plus at least $1M a year in O&M; an 8-hour version at $118-144M plus ~$360K O&M. The headline levelized-cost claims are just under $0.04/kWh unsubsidized today, with $0.01-0.02 claimed at roughly 100 GWh of annual production — about a million units a year, per TechCrunch’s November 2025 reporting. Abraham has said data centers would pay three times that. ExoRise shifts revenue toward development economics — land plus energy delivered as a package, PPA-style — which monetizes scarcity of powered sites rather than module margin. Nothing about actual contracted pricing has been disclosed, and no revenue figures exist publicly.

Traction over time

Read the backlog series skeptically: the unit of measure changed from megawatts (April 2024) to gigawatt-hours (late 2024), which makes the growth uncomparable, and Latitude Media reported the September 2024 backlog was primarily memoranda and letters of intent, not firm orders. The backlog has also been essentially flat at ~90 GWh from April 2025 through November 2025. The hardest number in the table is the quietest one: zero disclosed operating commercial megawatts as of mid-2026.

Market analysis

The demand side is not in dispute. The IEA (April 2025) projects global data-center electricity consumption roughly doubling from 485 TWh in 2025 to about 945 TWh by 2030; Goldman Sachs (2025) forecasts a 165% increase in data-center power demand by 2030 versus 2023 and ~122 GW of capacity online by decade’s end, with data centers driving almost half of US electricity demand growth. Multi-year interconnection queues and sold-out gas-turbine order books (GE Vernova’s slots stretch to 2028-29) make behind-the-meter, off-grid generation genuinely valuable — this is the structural force Exowatt surfs. The catch is Exowatt’s serviceable market is far smaller than the TAM: CSP-grade direct sunlight exists only in narrow Southwest desert bands. Abraham claims ~60% of US data centers (~750 facilities) have sufficient sun; Columbia’s Travis Bradford calls the realistic geography “pretty narrow bands,” in places that also fight heat and water constraints.

Competitive intel

Exowatt’s competitors sort into three tiers. The commodity stack — PV plus lithium batteries, integrated by the likes of Fluence — is the existential one: Lazard’s 2023 numbers already put unsubsidized utility-scale PV as low as $0.024/kWh, below Exowatt’s current claim, on a globally standardized supply chain. BNEF solar analyst Jenny Chase’s dismissal was total: photovoltaics can also power data centers. The thermal-battery peers — Antora ($272M raised, solid-carbon blocks, no-moving-parts thermophotovoltaic conversion) and Rondo ($107M+, refractory brick, heat-first) — chose to charge from cheap grid electricity rather than concentrate sunlight, which means they ride the falling PV cost curve instead of competing with it; Exowatt is the only one of the three that bet against PV at the point of capture. Rondo’s CIO John O’Donnell ran Ausra, the modular-CSP startup that died in 2014 — the industry’s institutional memory arguing against Exowatt’s architecture by career choice. For firm data-center power specifically, Fervo ($1B+ raised, Google contracts, ~500 MW Cape Station under construction) sells the same 24/7 promise with operating references and no sun constraint, and gas turbines remain what desperate hyperscalers actually buy. Exowatt’s genuine edges: factory-built modularity, an all-American supply chain in a tariff era, and speed-to-power on land it now controls via ExoRise.

History and evolution

• 2015-2022 — Happi (then Parvizian) founds and runs Volansi; it collapses and Sierra Nevada buys the assets, Oct 2022.
• Early 2023 — Abraham recruits Happi to Miami; Exowatt incubates inside Atomic; 50+ design iterations.
• Apr 2024 — Out of stealth via WSJ exclusive; $20M seed from a16z, Atomic, Sam Altman; claims 500+ MW of demand.
• Sep 2024 — P3 unveiled at a lavish LA airport-hangar party (DiCaprio front row, call-and-response chants) timed to RE+; 1.2 GW backlog claimed. Days later, Latitude Media publishes its “we’ve seen it before” teardown.
• Apr 2025 — $70M Series A led by Felicis — notably structured as $35M equity, $35M venture debt from HSBC. Deployments promised “beginning 2025.”
• Mid-2025 — First pilots with commercial customers reported; earlier reporting pointed to a West Texas crypto-mining site as the initial deployment target.
• Oct 2025 — Fast Company “Next Big Thing in Tech” award.
• Nov 2025 — $50M Series A extension (MVP Ventures, 8090 Industries); backlog ~90 GWh.
• Jan 2026 — ExoRise launched: from module vendor to powered-land developer.
• Mar 2026 — Austin engineering/manufacturing campus opens; first ExoRise pilot targeted for end of 2026.

What people say

The case for. Investors frame Exowatt as an American-dynamism industrial bet: a16z’s Katherine Boyle cites US manufacturing and energy resilience; Felicis backed the claim of minimal-degradation storage cheaper than lithium; 8090 Industries’ Rayyan Islam pitched it as 24/7 power for hyperscalers that “doesn’t have time” for the grid. Fast Company gave the P3 a 2025 Next Big Thing award; SiliconANGLE’s November 2025 profile was warm; and the sheer investor roster — Altman, DiCaprio, Felicis, Starwood, Thrive — plus a claimed 90 GWh pipeline says the demand story lands with sophisticated buyers of the narrative, if not yet of the product.

The complaints. The skeptics are unusually specific and unusually credentialed. Latitude Media’s September 2024 investigation assembled the case: BNEF’s Chase said solar-thermal combines the disadvantages of fossil fuels (moving parts, heat engines) with the disadvantages of PV (large collection area, variable resource). Energy Transition Ventures’ Craig Lawrence posed the killer question — CSP lost when PV and batteries cost ten times more than today; “if they couldn’t win back then, how do they win now?” Columbia’s Bradford called the geography “really really small” and the reliability match to five-nines data centers “a very tough match.” Scale Microgrids’ Duncan Campbell publicly asked if anyone else had “run the napkin math.” The component history is a graveyard: Stirling-engine solar companies (Stirling Energy Systems, Infinia) went bankrupt in 2011-13; modular-CSP Ausra shut in 2014; Heliogen, the last celebrity CSP venture, lost ~99% of its SPAC value. Observers compared the hangar-party theatrics to Bloom Energy’s 2010 “power plant in a box” reveal — and to WeWork. Employee-review data is effectively nonexistent (no meaningful Glassdoor presence at ~78 staff), which itself means the culture and execution claims are unaudited. And the backlog: units shifted mid-stream from MW to GWh, and reporting indicates it is mostly non-binding.

Outlook: the open question

For Exowatt to work, three things must prove true: the P3’s unsubsidized cost must actually land below ~$0.04/kWh in real deployments and fall toward $0.01-0.02 as Austin-built volume scales; the Stirling-engine fleet must run reliably enough, for years, that five-nines data centers accept it as primary power without a gas backstop that eats the economics; and ExoRise must convert the ~90 GWh of soft demand into binding contracts on Southwest land faster than hyperscalers’ alternatives — gas turbines, Fervo’s geothermal, or plain PV+BESS oversized with storage — can be delivered. The bull case is real: power scarcity is the binding constraint of the AI buildout, turbine order books are sold out for years, interconnection queues run half a decade, and a factory-built, US-supply-chain module that shows up on trucks and works day one has genuine option value; the pivot to selling powered land monetizes the scarcest asset (energized sites) rather than the most contested one (LCOE).

What would kill it is precisely what killed its ancestors: the PV+BESS cost curve. Every prior Stirling and modular-CSP venture died not because the physics failed but because commodity silicon and lithium got cheap faster than proprietary hardware could scale — and both are still getting cheaper, on supply chains of a scale no venture-backed manufacturer can replicate. If Exowatt’s first ExoRise site slips past end-2026, if measured LCOE at the pilot comes in materially above $0.04, or if the backlog stays non-binding through another funding cycle, the honest read will be that this was a well-marketed rerun. The tells to watch are unglamorous: commissioned megawatts, a named paying customer, measured cost per kWh at an operating site, and whether the next round is priced by an investor underwriting the machine rather than the story.

Sources and further reading

• The trouble with Exowatt’s ‘groundbreaking’ tech: We’ve seen it before (Latitude Media, Sep 2024)
• Sam Altman-backed Exowatt wants to power AI data centers with billions of hot rocks (TechCrunch, Nov 2025)
• Exowatt Closes $70 Million in Series A Funding to Deploy and Scale the Exowatt P3 (GlobeNewswire, Apr 2025)
• Exowatt Raises Additional $50 Million to Accelerate U.S. Rollout of Dispatchable Solar for the AI Era (Yahoo Finance/BusinessWire, Nov 2025)
• Sam Altman-backed Exowatt targets ‘Frontier Land’ data centers with modular solar thermal storage (pv magazine USA, Jul 2026)
• Exowatt is embracing a new business model (Latitude Media, Jan 2026)
• Exowatt Launches ‘ExoRise’ to Deliver Turnkey Powered Land and Energy for the AI Era (Exowatt, Jan 2026)
• Sierra Nevada buys drone maker Volansi after VC funding dries up (Defense News, Oct 2022)
• Exowatt aims to repurpose old tech to deal with new AI power demand (Canary Media, Sep 2024)
• AI to drive 165% increase in data center power demand by 2030 (Goldman Sachs Research, 2025)
• Energy and AI: energy demand from AI (IEA, Apr 2025)