GM’s Sodium-Ion Bet: Why It’s Selling Cells, Not Systems

The most interesting thing General Motors said this week was not about a battery. It was about where to stand. GM is moving into grid and data-center energy storage, one of the most contested hardware markets going right now, and it has decided not to build the storage systems. It is going to make the cells and sell them to the company that does. That reads like a smaller ambition than the headlines suggest. I think it is the more disciplined one, and it is the call I spend a good part of my week pushing founders to make.

What GM committed to

Start with the facts, because the shape of the deal matters more than the headline number. GM has put $900 million toward commercializing new battery chemistries, including its own development center, and is building sodium-ion cells with trial production targeted for 2028. It does not plan to sell finished storage systems. It plans to sell those cells to Peak Energy, a startup that integrates them into grid and data-center installations. In parallel it sells lithium iron phosphate cells to LG Energy Solution for interim systems, and through Redwood Materials it is routing retired EV battery packs into stationary microgrids, including a 12-megawatt system at a Crusoe data center in Nevada and a smaller install at one of GM’s own plants in Michigan.

The chemistry choice is the tell. TechCrunch reports the sodium-ion cells are cheaper, longer lasting, and less prone to overheating than the lithium that goes into an electric car, at the cost of being bigger and heavier for the same amount of stored energy. Peak says systems built on them can drop the cooling and fire-suppression hardware entirely. None of that helps you in a vehicle, where every kilogram counts. All of it helps you in a box that sits bolted to a concrete pad and never moves.

Why the chemistry is a product call, not a science one

Sodium-ion is not a better battery than lithium in the way a car buyer would measure. It holds less energy per kilogram, which is exactly why nobody is putting it in a sports car. For a battery parked next to a data center, weight is free. What costs you there is dollars per kilowatt-hour, how many charge cycles the cell survives, and whether it can sit safely next to a building full of servers without a fire-suppression budget. GM picked the chemistry whose weakness does not apply to the job and whose strengths do. That is the whole move.

Founders get this backwards constantly. They chase the best number on every line of the spec sheet, because a high spec feels like progress and a traded-away spec feels like a concession. But the application only rewards two or three of those numbers, and it quietly penalizes the rest. A battery optimized for energy density is the wrong battery for a stationary site, even though it looks better on paper. Decide the duty cycle first and it tells you which specs to fight for and which to give up without apology. The team that defines the job wins the spec argument before it starts.

Deciding which layer is the product

The harder discipline is the second decision: GM chose to be the cell, and let Peak own the system built around it. I learned how hard that call is to force years ago at Ascend Communications. We had something like 30 million lines of code running most of the world’s dial-up access, and the company, which thought of itself as a hardware company, could not see that the software was the product until somebody named it and drew a boundary around it. I pushed to call it an operating system and sell it as one. I did not write a line of that code; the engineering teams did. My job was to recognize which layer we actually sold and say so out loud.

That is the move GM is making here. It is not trying to own the cell, the enclosure, the controls software, and the field install all at once. It is going to be very good at one layer and sell into the companies that own the others. For a founder burning runway, that kind of narrowing is survival, not timidity. The instinct runs the other way, because owning more of the stack feels like capturing more of the value. Most of the time it just means you are mediocre at five things instead of excellent at one, and the customer can tell.

Dave Saunders is the founder of Base Reality Group and a Fractional CPO for hard-tech founders. He was a founder and operator at Galen Robotics, where the surgical-robotics platform earned FDA De Novo authorization in 2023, and he managed a 35-patent portfolio licensed from Johns Hopkins. He wrote Founders Who Finish and publishes The Build. More about Dave →