Meta Reserves Up to 1 GW of Noon Energy's Carbon-Oxygen Batteries, Betting a NASA Mars Chemistry Can Power AI Data Centers for 100 Hours
Meta signed a reservation deal for up to 100 GWh of ultra-long-duration storage from Noon Energy, a startup whose reversible fuel cell borrows chemistry from NASA's MOXIE Mars experiment.
Overview
Meta has agreed to reserve up to 1 gigawatt, or 100 gigawatt-hours, of ultra-long-duration energy storage from Palo Alto-based startup Noon Energy, in a deal that pairs a hyperscaler hungry for round-the-clock clean power with a battery chemistry originally engineered for Mars. The agreement, announced on April 21, begins with an initial 25 megawatt / 2.5 gigawatt-hour project targeted for completion in 2028, with the larger 1 GW / 100 GWh supply contract to follow if the first system performs.
The partnership is one of the largest publicly disclosed reservations of multi-day storage technology to date, and it expands Meta’s clean energy portfolio beyond the 6.6 GW of nuclear power deals the company announced earlier this year.
What We Know
According to the announcement carried by GlobeNewswire, Noon’s system uses modular, reversible solid oxide fuel cells capable of storing and discharging electricity for more than 100 hours at a stretch. The technology stores energy using abundant elements, specifically carbon and oxygen, rather than relatively scarce metals like lithium. That matters for a customer planning tens of gigawatt-hours of capacity, because multi-day storage built on lithium would compete directly with the batteries the world is installing in electric vehicles.
The specific chemistry is a reversible electrofuels battery. Cleantechnica reports that Noon’s cell splits CO2 into solid carbon, which becomes the storage medium, and oxygen, which is vented to the atmosphere. During discharge, the system pulls oxygen back in from ambient air and recombines it with the stored carbon to produce electricity. The same publication notes the chemistry is adapted from NASA’s MOXIE experiment, which rode aboard the Perseverance rover and demonstrated CO2-to-oxygen conversion at roughly 800 degrees Celsius on the Martian surface.
Performance claims are aggressive. Per Electrek, Noon announced in January 2026 that it had operated a containerized commercial-scale prototype for thousands of hours, with individual tests exceeding 200 hours of continuous discharge. The company claims its system offers roughly 50 times the energy capacity of conventional lithium-ion storage at a footprint 200 times smaller than flow batteries or pumped hydro, while using about 1 percent of the critical materials required by lithium-ion. Cleantechnica cites a round-trip efficiency of roughly 80 percent from Cyclotron Road analysis and a long-term cost target below $1 per kilowatt-hour of capacity.
On the commercial side, Cleantechnica reports that Noon Energy’s funding to date has come from a mix of government grants and venture capital, including an early ARPA-E award, a California Energy Commission CalSEED grant, a 2023 Series A led by Clean Energy Ventures, and an $8.76 million California Energy Commission grant tied to a Yolo County demonstration. Meta’s reservation is by far the largest commercial commitment the company has disclosed.
Nat Sahlstrom, Meta’s vice president of energy and sustainability, said in the announcement that “bringing data centers online faster requires rapid deployment of reliable energy sources,” and that the agreement “advances that goal with a storage technology that delivers grid resilience and firm power.” Noon co-founder and CEO Chris Graves said in the same release that “we’re partnering with a company that is actively securing stable power for the AI infrastructure of tomorrow, and Meta recognizes the promise in our 100+ hour ultra-long duration storage technology.”
What We Don’t Know
Neither company has disclosed a financial value for the reservation or the specific sites where the Noon systems will be deployed. The GlobeNewswire announcement describes the arrangement as a reservation rather than a firm offtake, and the full 1 GW order is contingent on the success of the initial 25 MW project. Pricing for the delivered systems, the levelized cost of storage Meta expects to achieve, and the share of Noon’s future capacity this reserves are not public.
Several technical questions also remain open. Noon’s 80 percent round-trip efficiency figure is cited from Cleantechnica’s summary of earlier third-party analysis rather than from audited commercial-scale operation. Electrek’s reporting confirms the January demonstration ran at a containerized scale, but stepping from a single container to a utility-scale 25 MW installation — and then to a gigawatt fleet — will test manufacturing yield, stack degradation over thousands of cycles, and the economics of operating solid oxide cells at high temperatures for years at a time. The 2028 target for the first Meta project leaves little room to iterate if the first units underperform.
Analysis
The deal reflects a shift in how the biggest AI operators are thinking about grid firmness. A hyperscaler can, in theory, buy enough solar and wind to offset its annual consumption on paper, but AI training clusters need power continuously, and the value of being able to ride through multiple cloudy, windless days is growing as data center load rises. The GlobeNewswire announcement frames the agreement as part of Meta’s effort to accelerate the next generation of AI infrastructure with 24/7 baseload clean energy.
Noon’s proposed configuration, described in Electrek, is to pair its fuel-cell battery with a layer of lithium-ion storage: lithium handles seconds-to-minutes demand swings while the carbon-oxygen system covers multi-day gaps. If that hybrid works at cost, it competes directly with the case for new gas peakers and with the thermal and iron-air systems other hyperscalers are backing, including the iron-air installation Google previously committed to in Minnesota.
For now, Meta’s commitment is a reservation, not a delivered megawatt. But the scale of the headline number — 100 GWh from a startup still scaling up from containerized prototypes — signals how aggressively the hyperscalers are willing to front-run unproven long-duration storage chemistries to keep their AI expansion on a clean-power trajectory.