AI Startups Automate Waste Sorting

As world leaders gather for COP31 in Türkiye, AI and robotics companies are using automated sorting to recover materials from waste. Less than 7% of materials entering the global economy come from recycled or reused sources, while municipal solid waste exceeds 2 billion tons per year and could nearly double by 2050. NVIDIA sustainability head Josh Parker says AI can identify recoverable materials and automate sorting. The company describes four startups in its Inception program using these systems.

Taiwan-based AIRECO uses vision AI and robotic arms, powered by NVIDIA Jetson Orin NX modules, to sort plastics at a pilot recycling line in Tainan. Its test line has recognition accuracy above 95%. The startup says its planned full production line next year aims to raise annual processing capacity from 18,000 to 48,000 tons and cut operating costs by 85%. Its finer sorting is intended to help more used plastic return to the same product type, such as bottles recycled into bottles.

U.S. startup EverestLabs AI has about 150 computer vision-guided robots operating at recycling facilities in North America and Australia. The company runs AI inference on NVIDIA RTX GPUs at the edge and trains models on NVIDIA data center GPUs using CUDA. It estimates its systems have identified about 40 billion objects, recycled roughly 25,000 tons of material and avoided about 30,000 metric tons of carbon dioxide emissions. EverestLabs is also shipping agentic AI tools that provide facility staff with operational data, and says it is exploring use in adjacent industries.

U.K. startup Greyparrot’s Analyzer systems use computer vision above conveyor belts to identify objects by material and brand, including in dusty or poorly lit conditions and at belt speeds up to 4 meters a second. More than 300 systems in over 20 countries have made over a trillion bounding box detections. Greyparrot says checking 15 tons of waste now takes six hours, down from 375 hours. Facilities report efficiency gains of 10% to 30%; at Amcor, the system helped increase food-grade plastic yield by 12%. Coca-Cola, Colgate-Palmolive and Kenvue also use its technology to measure packaging recycling.

Berlin-based reverse.fashion uses high-resolution computer vision to sort used garments by type, defects, color, size, brand and other attributes. Its system processes textiles moving at 3 meters a second through over a dozen vision AI models, optimized with NVIDIA TensorRT and running on NVIDIA Jetson modules. The company estimates the system sorts in one hour what previously took a week and is rolling it out to facilities in Europe. Europe generates nearly 7 million tons of used textiles annually; about a quarter are separately collected and sorted by hand into hundreds of categories.

As world leaders gather for COP31 in Türkiye, AI and robotics companies are using automated sorting to recover materials from waste. Less than 7% of materials entering the global economy come from recycled or reused sources, while municipal solid waste exceeds 2 billion tons per year and could nearly double by 2050. NVIDIA sustainability head Josh Parker says AI can identify recoverable materials and automate sorting. The company describes four startups in its Inception program using these systems.

Taiwan-based AIRECO uses vision AI and robotic arms, powered by NVIDIA Jetson Orin NX modules, to sort plastics at a pilot recycling line in Tainan. Its test line has recognition accuracy above 95%. The startup says its planned full production line next year aims to raise annual processing capacity from 18,000 to 48,000 tons and cut operating costs by 85%. Its finer sorting is intended to help more used plastic return to the same product type, such as bottles recycled into bottles.

U.S. startup EverestLabs AI has about 150 computer vision-guided robots operating at recycling facilities in North America and Australia. The company runs AI inference on NVIDIA RTX GPUs at the edge and trains models on NVIDIA data center GPUs using CUDA. It estimates its systems have identified about 40 billion objects, recycled roughly 25,000 tons of material and avoided about 30,000 metric tons of carbon dioxide emissions. EverestLabs is also shipping agentic AI tools that provide facility staff with operational data, and says it is exploring use in adjacent industries.

U.K. startup Greyparrot’s Analyzer systems use computer vision above conveyor belts to identify objects by material and brand, including in dusty or poorly lit conditions and at belt speeds up to 4 meters a second. More than 300 systems in over 20 countries have made over a trillion bounding box detections. Greyparrot says checking 15 tons of waste now takes six hours, down from 375 hours. Facilities report efficiency gains of 10% to 30%; at Amcor, the system helped increase food-grade plastic yield by 12%. Coca-Cola, Colgate-Palmolive and Kenvue also use its technology to measure packaging recycling.

Berlin-based reverse.fashion uses high-resolution computer vision to sort used garments by type, defects, color, size, brand and other attributes. Its system processes textiles moving at 3 meters a second through over a dozen vision AI models, optimized with NVIDIA TensorRT and running on NVIDIA Jetson modules. The company estimates the system sorts in one hour what previously took a week and is rolling it out to facilities in Europe. Europe generates nearly 7 million tons of used textiles annually; about a quarter are separately collected and sorted by hand into hundreds of categories.