The AI compute boom is about to hit a wall made of transformers and transmission lines, unless someone figures out how to power data centers without breaking the grid.
The Summary
- Redwood Materials built a modular data center in Nevada that runs on solar power and hundreds of second-life EV batteries, essentially bypassing the electric grid
- Founded by JB Straubel (Tesla's former CTO), the $6B startup is turning waste into infrastructure at the exact moment AI is making energy the new bottleneck
- The U.S. added 18.9 gigawatts of energy storage capacity last year—enough for 15-20 million homes—and second-life batteries could scale this faster and cheaper than manufacturing new ones
The Signal
Data centers now consume roughly 4% of U.S. electricity, and that number is accelerating as every tech company races to train frontier models. The grid was not built for this. Data centers are getting blocked in Virginia. Ireland put a moratorium on new builds in Dublin. Local utilities are starting to say no because adding more load means higher bills for everyone else or building new power plants nobody wants to fund.
Redwood Materials found the gap in the market: used EV batteries still hold 70-80% of their original capacity when they're retired from vehicles. That's enough juice to stabilize a microgrid or smooth out solar intermittency, but not enough for a car company to warranty them. So they sit in warehouses or get recycled for raw materials. Straubel, who spent years at Tesla watching this waste pile up, saw the obvious play.
"There's got to be some way we can take advantage of that and be one of the first to really scale it up."
The Nevada site pairs these second-life batteries with solar panels and powers a Crusoe data center. No new battery manufacturing. No transmission line upgrades. No rate hikes for the town next door. The economics work because:
- Second-life batteries cost a fraction of new grid-scale storage
- Solar is cheap but intermittent; batteries make it reliable
- Data centers need 24/7 power but can be flexible about *where* they get it
This is the new stack for AI infrastructure. Not "find a place with cheap power and plug in." But "build your own microgrid where land and sun are abundant, use batteries others consider trash, and never touch the grid." It decouples AI compute growth from political fights over utility rates.
The timing is everything. U.S. energy storage additions hit 18.9 gigawatts last year. That's more capacity added in one year than the entire installed base a few years ago. California is leading. Texas is close behind. The storage boom is here, and it's being driven by two forces: renewable intermittency and data center demand.
Redwood is betting that second-life batteries can ride both waves. They're already recycling EV batteries at scale. Now they're proving those same batteries can power the infrastructure for Web4 without making enemies of every utility commission in America. The value proposition is simple: you get cheap, fast-deploy energy storage. The grid gets relief. The batteries get a second life instead of immediate disassembly.
The Implication
If this model works at scale, it changes the geography of AI. Data centers stop clustering near cheap grid power and start showing up wherever there's land, sun, and access to used batteries. That could mean more compute capacity, faster, without the political gridlock that's slowing builds in places like Northern Virginia.
Watch for other battery recyclers and energy storage companies to copy this playbook. The bottleneck for AI is increasingly energy, not chips. Whoever solves energy distribution without fighting utilities or ratepayers wins the next phase of the agent economy.