The grid can't keep up with AI's appetite, so one of Tesla's original engineers is turning dead EV batteries into a workaround worth billions.
The Summary
- Redwood Materials is pivoting from pure battery recycling to building "behind-the-meter" power systems that combine second-life EV batteries with solar panels, letting AI data centers bypass grid connection waits that now stretch for years
- Getting electricity to run huge AI data centers has become a major bottleneck for developers, while Microsoft is simultaneously preparing to install gigawatts of its own chips, intensifying the power crunch
- Redwood is already executing with Rivian and has built a system for Crusoe Energy in Nevada, proving the model works at scale
The Signal
JB Straubel, Tesla's co-founder who left to start Redwood Materials in 2017, bet on battery recycling as the next big infrastructure play. Now he's found an even bigger customer than EVs: data centers that can't wait 3-5 years for utilities to run them a power line. The math is simple. Every new AI training cluster needs hundreds of megawatts. The US grid wasn't built for this load, and utilities move at the speed of regulatory approval. Redwood's pitch is to make the grid optional.
The company's "behind-the-meter" systems work like this: retired EV batteries that still hold 70-80% capacity get repurposed as stationary storage, paired with solar arrays and potentially natural gas generators. The customer generates power on-site, stores it in the battery banks, and never touches the public grid. No utility approval. No interconnection queue. No three-year wait while your competitors scale.
"Connecting to the grid can take years."
Redwood has already installed a 10-megawatt-hour system at Rivian's Illinois plant and built another for Crusoe Energy in Nevada. Crusoe, for context, specializes in turning stranded energy into compute. They already run data centers on flared natural gas at oil fields. Partnering with Redwood extends that model: energy independence as a competitive moat.
The timing is perfect because the power crunch is hitting everyone simultaneously. Microsoft is gearing up to deploy gigawatts of its own custom AI chips, which means even more demand from one of the handful of hyperscalers who actually have capital to build. When Microsoft, Google, Amazon, and Meta are all competing for the same limited grid capacity in the same handful of regions with cheap power and fiber, someone has to lose. Or someone has to build different.
Key advantages of the Redwood model:
- Speed: bypass years of utility approval and interconnection studies
- Control: no exposure to rate hikes, grid instability, or regional power politics
- Sustainability optics: "powered by recycled batteries and solar" is better PR than "powered by natural gas peaker plants"
But here's the constraint: this only works if you can source enough second-life batteries. Redwood recycles Rivian's batteries and has partnerships across the EV industry, but the supply is still ramping. Most EVs on the road are less than five years old. The real flood of used batteries won't hit until 2028-2030. Redwood is building a business model that depends on a supply chain that doesn't fully exist yet.
The Implication
If Redwood's model scales, data center location strategy changes entirely. You don't need to build in Oregon or Iowa because that's where the grid has slack. You can build closer to users, closer to fiber, or in tax-friendly jurisdictions, as long as you have land for solar and a deal with a battery supplier. That fragments the AI infrastructure map and makes energy independence a competitive advantage, not just a cost center.
For anyone building AI companies: watch where the hyperscalers put their next data centers. If they start going off-grid with Redwood-style systems, it means the bottleneck has shifted from compute to power, and the power problem is being solved with capital, not patience. If you're raising money to train models, your investors will ask where you're getting the electricity before they ask about your architecture.