Skip to content
LIVE
The Executives BriefThe Executives Brief

Redwood's AI sorts old batteries at 98% accuracy, and the prize is lithium

JB Straubel's recycling startup built the system in a few months and now runs several tons an hour, turning a messy manual bottleneck into a proprietary data moat.

ByLama Al-RashidTechnology Correspondent, The Executives Brief
·4 min read
Redwood's AI sorts old batteries at 98% accuracy, and the prize is lithium
Executive summary

Redwood Materials, the battery recycling startup founded by Tesla cofounder JB Straubel, has built an automated AI sorting system that identifies batteries with more than 98% accuracy, according to the company, and can handle several tons an hour. For operators and investors, it is a case study in where AI value may actually land: unglamorous industrial bottlenecks where proprietary data, not model novelty, is the edge.

The most interesting AI deployment of the week is not a chatbot. Redwood Materials, the battery recycling company founded by Tesla cofounder JB Straubel, has built an automated sorting system that identifies batteries with more than 98% accuracy, according to the company. It was built in just a few months, and it can handle several tons of batteries an hour. That combination, high accuracy, fast build, real throughput, is the part worth pausing on. Recycling batteries sounds like an obvious win until you remember that someone has to sort through mountains of them first, and that job has historically been slow, manual, and expensive. Redwood's answer is to let machines do the looking.

Here is how the system works. Old batteries arrive in a jumble of boxes and bins. Machines separate them into a single-file line. Each battery then passes through a tunnel equipped with cameras and sensors that examine its label, shape, and weight. The AI figures out what kind of battery it is and sends it to the right place for recycling. Redwood developed the model itself, trained on the company's voluminous battery data, and it runs on a cluster of GPUs on site. The inference happens at the plant, not in a distant data center. The company owns the data, the hardware, and the process.

The sorting problem is harder than it sounds. Batteries come in all shapes and sizes, with different chemical ingredients. Mixing up the wrong ones can make recycling less efficient and more expensive. Get it wrong at scale and you contaminate a batch, waste energy, and damage the economics of the whole line. That is why sorting has typically been the least glamorous and most stubborn step in the recycling chain. It is also why the accuracy figure matters: at several tons an hour, small accuracy gains compound into real money, and a system that misroutes feedstock quietly bleeds value all day.

Why go to all this trouble? Because the payoff is material, literally. Redwood wants to recover valuable materials such as lithium, nickel, and cobalt rather than relying so heavily on mining and imports. Those three metals are the backbone of lithium-ion batteries, and their supply chains have been a persistent strategic headache for anyone building electric vehicles, grid storage, or consumer electronics. For years, the US has relied heavily on imports for battery materials and processing, which leaves domestic manufacturers exposed to price swings and geopolitical friction far outside their control. Domestic recycling offers a partial answer: instead of digging new material out of the ground, you pull it back out of the batteries already in circulation. But that only works if you can sort the feedstock cheaply and reliably. AI is the unlock, not the product.

This is a neat example of AI tackling a messy, unglamorous industrial problem, and potentially making American manufacturing more competitive. Note what is not here: no foundation model arms race, no consumer app, no viral demo. Just cameras, sensors, a trained model, and GPUs sitting on a factory floor. The strategic insight is that the model is only as good as the data underneath it. Redwood trained its system on its own battery data, which means the accuracy is not something a competitor can simply buy off the shelf. The moat is the dataset plus the physical plant, and both take years to assemble.

That has implications well beyond recycling. Every industrial operator sitting on decades of process data, inspection logs, and scrap rates is now holding a potential AI asset. The playbook Redwood demonstrates is narrow and repeatable: find a high-volume, repetitive, judgment-heavy task, instrument it with cheap sensors, train on your own proprietary data, and run inference locally where latency and data control matter. The build time, a few months, is the headline for anyone who has been told AI transformation takes years. It does not, when the problem is specific and the data already exists.

For executives, the questions are uncomfortable and useful. What is your sorting problem, the bottleneck everyone accepts as the cost of doing business? Do you own the data that would let a model solve it, or does a supplier? And if you solved it, would the savings show up in margin, in supply security, or in the ability to bid for contracts competitors cannot touch? Redwood's system is a reminder that the most defensible AI wins may be the ones nobody outside the plant ever sees.

Watch two things from here. First, whether Redwood publishes or verifies the accuracy figure independently; the 98% number is the company's own, and third-party validation would harden the claim. Second, whether the approach spreads to other messy feedstocks, from e-waste to scrap metal. If it does, the AI story of the next few years may be less about what models can say and more about what machines can sort.

Executive ActionsLocked

This story's Key Insights and Take-aways are locked.

Create a free account to unlock Executive Actions for one credit.

Register to Unlock

Always free for Executives Club members. Join the Club

More in Technology