65 GWh of Sodium-Ion Got Signed This Year. The Storage Map Follows.

When the world's largest battery maker signs 65 gigawatt-hours of sodium-ion deals in a single year, the chemistry stops being a research question and becomes a delivery schedule.

CATL built that schedule in 2026. In April the company signed a 60 GWh supply agreement with Chinese integrator HyperStrong, the largest sodium order ever placed, according to the announcement. Earlier this year CATL added a 5 GWh memorandum of understanding with Dutch energy-storage specialist Alfen, with deployments set to begin in 2027 (CATL-Alfen sodium-ion deal coverage).

CATL rates the cells for 15,000 charge-discharge cycles and a 25-to-30-year service life before they drop to 70% of original capacity, the company says. The cells hold more than 92% capacity at -20°C and survive 10,000+ cycles at 45°C, which addresses two failure modes that have kept lithium-ion expensive in extreme climates. Alfen has more than 1 GWh of storage already installed across Europe and has been buying lithium cells from CATL since 2023, so the new sodium line slots into an existing procurement relationship rather than a speculative one.

The US side of the curve is moving too. Peak Energy has shipped the country's first grid-scale sodium-ion battery, and GM has backed Peak's sodium platform in 2026, according to the trade coverage of the US sodium-storage buildout. Sodium is roughly a thousand times more abundant than lithium and far cheaper to source, the structural argument that has kept the chemistry in research labs for two decades.

The cost claim that has done the most work in sodium's public image is a different matter. Clean-tech commentator Ramez Naam, on a recent episode of the Moonshots podcast, called sodium-ion "10x cheaper" than lithium-ion (Moonshots EP #280). Sodium cells are presently more expensive per kilowatt-hour than mass-produced lithium iron phosphate at the cell level. The cost argument rests on raw-material price, cycle life, and the elimination of certain battery-management overhead, not on a ten-times cell-cost gap. A reader who wants to use "10x cheaper" should source it independently or drop it.

Sodium is the best-evidenced of the three cheap-power bets the same guest laid out on that episode. The other two are worth flagging because they are circulating in the same energy-and-AI conversation, even if the evidence base is thinner.

The first is grid headroom. Naam argues the United States may already have, sitting mostly unused, enough transmission capacity to electrify a large slice of cars and heating. The 200-gigawatt figure is his, not a published grid study. If the underlying interconnection data backs it up, the implication is that the binding constraint on the energy transition is permitting and dispatch, not steel. Until that data is sourced, the figure should be read as a guest estimate, not a measurement.

The second is wave-powered data centers. Oregon-based Panthalassa raised $140 million this year, led by Peter Thiel, to build floating wave-power orbs paired with onsite AI compute, transmitting data back to shore via low-earth-orbit satellites rather than subsea cables (GeekWire on Panthalassa). The pitch is that an AI inference cluster can be sited where the wave resource is strongest instead of where the grid is. The round is real; the unit economics are still a 2027 story.

The thread tying all three bets together is the demand side. AI training and inference have moved power from a side concern to a site-selection variable. Sodium has a procurement map, grid headroom has a queue, wave-powered compute has a satellite backhaul. The 2026 signal is that all three are being tried, separately, by people who think the bottleneck on AI-era power is going to be physical rather than financial.

The CATL-Alfen 5 GWh MOU converts into a binding order before year-end. Peak Energy's GM-backed platform lands a US utility offtake. Either would turn 2026 from a sodium procurement milestone into the start of a storage map reset.