In November 2021, I examined Wright Electric’s proposal to power passenger aircraft with aluminum–air batteries. Electric aviation was drawing extraordinary amounts of speculative capital. Within weeks, I calculated that five publicly traded urban-air-mobility companies had lost about $16 billion in market capitalization from their 2021 peaks. Wright wasn’t one of those listed companies, but its aluminum proposal belonged to the era’s search for exciting new energy pathways. Five years later, aluminum–air has found a different bubble to float in: AI data centres.
California startup Voya Energy announced a $35 million funding round in August 2026 to develop generators that turn scrap aluminum pellets, air and water into electricity. There’s no combustion and no exhaust at the site. The company hopes to displace diesel backup generators and also pitches power for data centres facing grid constraints. The electrochemistry is real, and better control of corrosion and spent electrolyte would be useful engineering. The economics and materials logistics deserve much more skepticism than the funding announcement invites.
An aluminum–air battery is mechanically refuelled, not electrically recharged like a lithium-ion battery. The aluminum is consumed, combining with oxygen and water to produce aluminum hydroxide. Metallic aluminum contains stored energy because a great deal of electricity was previously used to separate it from oxygen. Recovering that metal from hydroxide requires processing and another substantial electricity input. Aluminum isn’t a primary energy source; it’s an exceptionally roundabout energy carrier.
That was my central criticism of the aviation proposal. Smelters and recycling facilities are located around industrial energy and materials infrastructure, not airports. Prepared metal would have to travel to aircraft, while spent material would have to travel back to processors. Wright and Columbia University subsequently received ARPA-E support for aluminum–air flow-battery work, so the chemistry hasn’t vanished. In a further irony, Wright now advertises a containerized data-centre generator, but that is a separate turbine-based product running on diesel or jet fuel, not an aluminum–air generator.
I also missed a particularly ugly aircraft constraint in 2021: it would get heavier as it flew if the reaction products stayed aboard. Each tonne of aluminum consumed produces about 2.9 tonnes of aluminum hydroxide, incorporating water that the aircraft would have carried and oxygen drawn from the air. With onboard water and retained hydroxide, the net aircraft mass increases about 890 kilograms per tonne of aluminum consumed. A flight using five tonnes would land around 4.4 tonnes heavier, while a conventional jet gets lighter as it burns fuel. The accumulating hydroxide could also form a slurry or solid that complicates storage and centre-of-gravity management. That’s a miserable feature for an aircraft fuel system.
The broader material problem is worse now because the world urgently needs to recover more metals. Recycling metallic aluminum takes about 95% less energy than primary production. Sorting, separating and remelting more scrap reduces the demand for new mining, refining and smelting. The same imperative applies to copper, steel and other materials needed for electrification. Using metallic scrap as fuel deliberately destroys the very chemical state that makes conventional recycling so valuable.
Voya says it will initially use low-grade, contaminated scrap that is difficult to recycle into high-quality products. That’s a more serious case than feeding beverage cans to a generator. But difficult to recycle is not synonymous with worthless or unrecoverable. Low-grade aluminum has competing uses, and improved sorting and alloy separation are part of industrial decarbonization. Creating a new demand for recovered metal as an expendable fuel competes with that priority. This is thermodynamic vandalism with a very long supply chain.
The proposed eventual closed loop doesn’t rescue it. An April 2026 lifecycle study modeled aluminum–air electricity storage at 29–35% electricity-to-electricity round-trip efficiency, depending on the regeneration pathway. Roughly three units of electricity have to go in for one to come back out, with material processing and transport along the way. Rechargeable batteries return far more of their input electricity. Density at the destination is useful, but it doesn’t cancel the losses at the smelter.
Consider a data centre requiring 100 MW continuously. Using Voya’s reported 57% conversion efficiency and aluminum’s roughly 8.1 kWh/kg theoretical electrochemical energy, it would consume about 520 tonnes of aluminum every day and produce about 1,500 tonnes of aluminum hydroxide. That’s roughly 190,000 tonnes of metal and 550,000 tonnes of byproduct each year, before considering the facilities, equipment and transport needed to manage the flow. The calculation illustrates continuous generation, not Voya’s proposed initial backup-power duty cycle. At occasional backup operating hours, the material flows would be much smaller.
Data-centre developers choose locations around fiber networks, land, grid access, customers and development economics. Smelting and aluminum recycling take place in different industrial geographies. A data centre beside an existing integrated aluminum-processing facility, with genuinely stranded scrap and a nearby byproduct customer, might have a niche case. A fleet of remote data centres importing hundreds of tonnes of metal and exporting even more chemical residue every day is an altogether different proposition. Colocation can shorten truck routes. It cannot fix the round-trip energy balance.
The price claim needs the same scrutiny. In a 2025 submission to the California Energy Commission, Voya listed targets of less than 30 US cents per kWh for electricity and 50 cents per pound for metal fuel. At 57% conversion efficiency, that assumed aluminum price implies roughly 24 cents per kWh for metal alone, before preparation, equipment, handling, logistics, financing and profits. Byproduct sales could provide a credit, but the revenue depends on purity, processing costs and customers for the hydroxide. Selling the hydroxide into industry and sending it back for metal regeneration are different routes, not simultaneous credits. Thirty cents is a target, not demonstrated commercial generation economics.
Diesel backup replacement still deserves a fair hearing. A generator that runs only during outages, avoids local exhaust and uses a compact storable fuel could be useful where diesel permitting or emissions are unusually burdensome. The lower operating hours also make a world of difference to the material quantities. But presenting aluminum as a geographically portable route to continuous electricity for AI data centres invites a completely different industrial and economic assessment.
Five years ago aluminum–air was proposed for aircraft during a boom in speculative aviation capital. Today the pitch has landed in AI infrastructure. The world needs more electrical generation and better-connected grids, while also recovering far more valuable metals. Consuming recovered aluminum to avoid building electricity infrastructure works against both objectives. The electrochemistry can be clever while the system it enables remains deeply misguided.
If you want energy analysis that follows the chemistry, material flows and money rather than funding rounds, subscribe to TFIE Strategy Briefing.