Unconfirmed: Nvidia's blog says Taiwan's AIRECO aims to cut operating costs 85% on a production line planned for next year, a target Automatica has not confirmed.
The Nov. 11, 2026 post, profiling four Inception startups1 that use vision AI to sort waste, puts that line in 2027. It says the AIRECO team "aims to nearly triple annual waste processing capacity from 18,000 to 48,000 tons while cutting operating costs by 85%." The post opens by referring to world leaders gathering for COP31 in Türkiye. It cites no audit, baseline, methodology or independent verification for the target. Automatica has not seen independent data, and we have no earlier reporting on AIRECO to set against it.
The missing baseline matters. The post does not say whether the 85% refers to total operating costs or cost per ton, or what it is measured against. Capacity of 48,000 tons is 2.7 times the 18,000-ton figure. On our arithmetic, if total costs fell 85% while throughput rose that much, cost per ton would fall by roughly 94%. If the 85% is per ton, total costs would fall by about 60%. The two readings imply very different plants.
The post also says AIRECO runs inference on Nvidia Jetson Orin NX modules on a pilot line at a recycling plant in Tainan, including to drive robotic arms, and that recognition accuracy on the test line "tops 95%." That figure is in the post's own voice, with no test conditions given. Chun-Feng Liu, AIRECO's founder and CEO, is quoted on why finer sorting enables closed-loop recycling, but not on the cost target.
The other three profiles follow the same pattern. The post says EverestLabs AI has about 150 robots deployed across North America and Australia. It attributes to the company estimates of about 40 billion objects identified, roughly 25,000 tons recycled and about 30,000 metric tons of carbon dioxide avoided. The post says Greyparrot's more than 300 Analyzer systems in over 20 countries have made over a trillion bounding box detections. It says they cut the time to check 15 tons of waste from 375 hours to six, and that Amcor raised food-grade plastic yield by 12%. For reverse.fashion, the post says Karsten Pufahl, co-founder and CTO, estimates the system sorts in an hour what once took a week.
None of those figures comes with a disclosed method in the post. The post is also first-party. Nvidia sells the Jetson modules and GPUs the startups use, and Josh Parker, its head of sustainability, is quoted saying the company is "giving innovators the computing tools" to expand the work. A vendor profile of customers establishes what Nvidia chose to publish, not how the systems perform.
What would confirm the AIRECO claim is audited results from the production line once it is running, with the cost baseline stated. What would undercut it is a baseline showing the 85% measures something narrower than overall operating cost. For the other firms, independent tonnage and detection data would test the post's figures. Neither Nvidia nor AIRECO has said more in anything Automatica has seen beyond the blog itself.