Innovation & Tech

Redwood Materials' AI Sorts Old Batteries at 98% Accuracy

Redwood Materials, founded by Tesla's JB Straubel, built an AI vision system in months that identifies batteries with 98%+ accuracy and sorts several tons per hour.

By Grace Kim

2 min read

Redwood Materials found a very practical use for AI: Sorting old batteries
Redwood Materials found a very practical use for AI: Sorting old batteriesanalogicus / pixabay

What's News

  • Redwood Materials' AI sorting system identifies batteries with more than 98% accuracy, per the company.
  • The system was built in just a few months and processes several tons of batteries an hour.
  • The AI model runs on an on-site GPU cluster and was trained on Redwood's own battery data.
  • Redwood was founded by Tesla cofounder JB Straubel.
  • The goal is recovering lithium, nickel and cobalt to reduce reliance on mining and imports.

Redwood Materials says its new AI-powered sorting system identifies old batteries with more than 98% accuracy — and it can process several tons of them every hour.

The battery recycling startup, founded by Tesla cofounder JB Straubel, built the automated system in just a few months, according to the company. It tackles one of the least glamorous problems in the recycling business: before batteries can be recycled, somebody has to sort through mountains of them first.

By Grace Kim

Why does battery sorting need AI?

Batteries arrive at recycling facilities in a jumble of boxes and bins. They come in all shapes and sizes, and they carry different chemical ingredients. Mixing up the wrong ones makes recycling less efficient and more expensive.

That mismatch is the core problem. A recycling plant that can't tell a lithium chemistry from a nickel one ends up with contaminated material streams, and that costs money on every ton processed.

Redwood's answer is a computer vision pipeline the company detailed publicly. Machines first separate the incoming jumble into a single-file line. Each battery then passes through a tunnel equipped with cameras and sensors that examine three things:

  • its label,
  • its shape,
  • and its weight.

From that data, the AI model determines what kind of battery it is and routes it to the right place for recycling.

What's under the hood?

Redwood developed the AI model itself and trained it on the company's voluminous battery data — an asset built from years of handling used cells. The model runs on a cluster of GPUs on site, not in the cloud, which fits the industrial scale of the operation.

The build speed is the other headline figure. The company says the entire system came together in just a few months, and it now handles several tons of batteries an hour.

For a hardware-plus-software system operating at that throughput, a few months from concept to production line is fast by any manufacturing standard.

What does this mean for US manufacturing?

The economics point at materials. Redwood wants to recover valuable metals such as lithium, nickel, and cobalt from the used batteries it processes. Every ton recovered domestically is a ton the US doesn't have to source through mining or imports — a supply chain concern that has grown alongside the electric vehicle and data center booms.

The company's timing aligns with surging demand for battery materials, including recycled EV batteries being tapped to power AI data centers.

It's also a useful data point in the debate over where AI delivers practical value. Rather than a consumer-facing feature, Redwood applied machine learning to a messy, unglamorous industrial problem — sorting scrap — where a 98% identification accuracy rate translates directly into processing efficiency and cost.

If the approach scales, it strengthens the case that AI's most durable wins in manufacturing will come from exactly this kind of unglamorous, high-volume sorting and quality-control work.

Source: feeds.businessinsider.com

Share this article:

More from Grace Kim

Grace Kim

Show full bio

Market editor covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.

2,582 articles

Related articles

« Previous article