Loading...
Carbon Robotics’ story began not in a lab, but over lunch in Idaho. Founder Paul Mikesell asked a farmer about his biggest challenges. The answer was simple — and ancient: weed control.
That conversation led to an audacious solution. The company’s LaserWeeder G2 spans 20 feet and carries 12 modules, each powered by two NVIDIA GPUs — 24 in total. This compute muscle lets it identify and incinerate up to 10,000 weeds per minute, without chemicals.
In an age of herbicide-resistant plants, it’s a game-changer. “There’s no such thing as a laser-resistant weed,” says Alex Sergeev, chief technology officer of Carbon Robotics.
Every LaserWeeder doubles as a data-gathering machine. Images from the field feed into a custom labeling tool, building what the team believes is the world’s largest labeled agricultural image dataset: over 65 million images, fueling the company’s “large plant model,” a play on the large language models that power chatbots like ChatGPT.
When GPUs finish their tour on the farm, they head to Carbon’s “retirement facility” — a Seattle-based data center — where they’re used to train the next generation of models. Optimized with NVIDIA TensorRT, the resulting foundational model works across all crops, creating a durable competitive edge.
From there, Carbon turned to another urgent challenge: a global shortage of tractor drivers. Over 25% of edible U.S. crops go unharvested due to labor gaps. The answer: the Carbon AutoTractor, an autonomous retrofit for existing machines. Built with farmers’ input, it runs around the clock, monitored by a remote operations center that can take control if, say, a deer wanders into the field.
Since its founding in 2018, Carbon Robotics has shipped more than 150 LaserWeeders to farmers in 14 countries, eliminating over 30 billion weeds. With investment from NVentures, NVIDIA’s venture arm, this Pacific Northwest company is proving that physical AI can tackle some of humanity’s oldest problems — from the ground up.
