NVIDIA: Vision AI Helps Recyclers Sort Plastics and Textiles

NVIDIA has detailed how four startups in its Inception program for startups are applying vision AI and robotics to sorting mixed waste, describing deployments that range from plastic recovery in Taiwan to textile grading in Europe. NVIDIA frames the work around a circular-economy challenge: globally, less than 7% of materials entering the economy come from recycled or reused sources, and municipal solid waste generation accounts for over 2 billion tons of trash per year. The startups each occupy a different part of the sorting problem, from robotic picking on material recovery lines to measurement of packaging and garment quality assessment.

Key Takeaways

• Four NVIDIA Inception members are described as using vision AI for sorting or measurement: AIRECO, EverestLabs AI, Greyparrot and reverse.fashion.
• AIRECO reports recognition accuracy above 95% on its pilot line at a recycling plant in Tainan, running inference on NVIDIA Jetson Orin NX modules to drive robotic picking arms.
• EverestLabs AI says about 150 robots are deployed in facilities across North America and Australia, and estimates its systems have identified about 40 billion objects and recycled roughly 25,000 tons of material.
• Greyparrot reports more than 300 Analyzer systems in over 20 countries and says facilities using its real-time data report efficiency gains of 10% to 30%.
• reverse.fashion says its system sorts in an hour what previously took a week, and is currently rolling out to facilities in Europe.

Context & Analysis

The NVIDIA post describes a common technical pattern across the four companies: models trained on NVIDIA GPUs, often in data centers or workstations, then run for inference at the edge on NVIDIA Jetson modules or RTX GPUs close to the conveyor belt. That split matters because sorting decisions must happen while material is moving. EverestLabs AI chief executive R. Paul Singh, paraphrased in the post, contrasted recycling robots with other robots that have unlimited time to complete tasks, noting that conveyor belts arrive at a set speed and require action before material moves on.

AIRECO, based in Taiwan, focuses on plastics. The company says it uses vision AI to sort plastics more finely so material can be pulled from mixed scrap piles for closed-loop recycling, in which used items are reprocessed into new versions of the same product. Founder and CEO Chun-Feng Liu, paraphrased in the post, argued that finer sorting increases the likelihood an item stays in the loop, such as a bottle recycled back into a bottle rather than into harder-to-recycle products. On the test line, recognition accuracy tops 95%. In a full production line planned for next year, the team aims to nearly triple annual waste processing capacity from 18,000 to 48,000 tons while cutting operating costs by 85%. Those capacity and cost figures are stated as targets for a planned line, not achieved results.

Competitive Landscape

The evidence supports comparison only among these four documented participants and their described roles; it does not establish market share, pricing or relative performance against unlisted vendors.

Two of the four, AIRECO and EverestLabs AI, describe robotic picking; Greyparrot describes measurement above the belt, and Amcor is cited as a packaging company using its data, with a 12% increase in yield of food-grade plastics. reverse.fashion identifies and categorizes used garments. Brands including Coca-Cola, Colgate-Palmolive and Kenvue are cited as using Greyparrot technology to measure how their packaging is recycled in real sorting plants.

What It Means

For Enterprise Buyers

Recycling facility operators, packaging producers and textile handlers evaluating vision AI should note that the figures in the NVIDIA post are company-reported and tied to specific lines, sites or time periods. AIRECO's 95% accuracy applies to a pilot line, while its tripling of capacity and 85% cost reduction are aims for a planned full production line. Buyers would reasonably ask for site-specific validation, throughput under their own belt speeds and material mixes, and clarity on integration with existing sorting equipment before treating vendor estimates as procurement benchmarks.

For Investors

The post offers deployment indicators rather than audited financials: robot counts, system counts, country reach and company estimates of material recycled or emissions avoided. Those signals can inform diligence on commercial traction, but investors should separately verify revenue, contract values, margins and customer retention before drawing valuation conclusions. The shared dependence on NVIDIA hardware and software is a documented technical commonality, not evidence of exclusivity or of commercial outcomes for NVIDIA.

Forward Outlook

Documented milestones to watch include AIRECO's planned full production line next year, with its stated capacity and cost targets; reverse.fashion's rollout to European facilities; and EverestLabs AI's stated work on classifying and picking every kind of object while shipping agentic AI tools for facility staff and looking to extend that capability to adjacent industries. Greyparrot's cited customer results at Amcor and brand users provide reference points for measuring packaging data in real plants. Risks and unanswered questions include whether pilot accuracy holds at production scale, how performance varies with waste mix and belt speed, and how facility efficiency gains attributed by customers are measured. Buyers should seek independent validation, pricing and commercial terms before making commitments.

Direct source: NVIDIA blog post.