Four Workers Still Stand Next to China's Fastest Sorter
A human worker in Zhangjiagang can sort about 100 kilograms of used clothing per hour. The Fastsort-Textile can handle that same 100 kilograms in two to three minutes. The math makes the person look slow. The story is not about that.
The story is about the four workers per shift standing next to the machine.
China generates roughly 36.4 million tons of textile waste per year, about a quarter of the global total, according to DontWasteIt. Only about 5.15 million tons get recycled. The sorting bottleneck is the problem. Over 80 percent of textile sorting in China still relies on manual labor, and a skilled worker can only sort around 100 kilograms per hour. Labor costs eat up more than 40 percent of operating expenses in the traditional model.
DataBeyond, a Chinese AI robotics company founded in September 2018, built the Fastsort-Textile to solve exactly that. The machine uses hyperspectral recognition — an AI that reads the chemical composition of fabric in milliseconds — to sort clothing by material type as it moves along a conveyor belt at 4 meters per second. High-pressure air nozzles eject the target material with what the company says is over 98 percent accuracy. The full automated line processes approximately 2 tons per hour, according to DataBeyond's own figures.
The system went into Shanhesheng Environmental Technology Ltd., a textile recycling facility in Zhangjiagang, in 2025. It was named to Time magazine's Best Inventions of 2025 list in the Special Mentions category.
Here is the part the headlines skip: the full automated line runs with four workers per shift. Not zero. Not 30. Four. The original manual operation at Shanhesheng needed more than 30 workers to process roughly 15 tons of post-consumer textiles in a single 8-hour shift. The new line delivers the same throughput with a fraction of that headcount. But the workers did not disappear. They moved from sorting to supervising.
"We moved from doing the task to managing the machine," one Shanhesheng employee said in comments provided to ABC News. "You have to understand, this job was not easy. It was physically hard, the conditions were not good. Now we are monitoring the line."
The distinction matters. Automation coverage defaults to a binary: robots take jobs, or they do not. The Fastsort-Textile suggests a third option that is harder to write about and harder to quantify — a reskilling that preserves the employment while transforming the nature of the work. The four remaining workers per shift are not doing the same job at lower volume. They are doing a fundamentally different job: watching for jams, handling exceptions the AI flags as ambiguous, managing the flow of material that confuses the hyperspectral sensor.
DataBeyond and Shanhesheng say the machine reduced the proportion of processed textiles deemed unrecyclable from 50 percent to 30 percent — a significant drop in material that would otherwise go to landfill or incineration. Shortly after commissioning, Shanhesheng received a 200-ton order from a global apparel company, a commercial signal that the high-purity sorted output has buyers.
That order is worth noting. Textile recycling at scale only works if the sorted material is clean enough to re-enter the supply chain. The Fastsort-Textile's claimed 98 percent accuracy is the mechanism that makes that possible — and the reason DataBeyond has attracted investment from Sequoia Capital China, Amber Capital, and Shengqu Capital.
There is a flag on this story, and it belongs in the article. Every throughput number, accuracy figure, and efficiency comparison in the available public record traces back to Shanhesheng's own internal analysis or DataBeyond's company announcements. No independent auditor has verified the 2 tons per hour processing rate, the 98 percent accuracy claim, or the before-and-after unrecyclable rates. These are the company's numbers on the company's letterhead. The deployment is real — the machines are running — but the performance claims rest on internal analysis that outside researchers have not confirmed.
What the article can verify: the system is deployed, it is running, it was named a Time Best Invention, and the four-workers-per-shift model is a documented feature of this specific implementation. Whether those four roles are genuinely better jobs than the manual sorting they replaced is a question that deserves follow-up reporting — the workers quoted in available coverage were generally positive, but a small sample from a single facility is not a labor policy conclusion.
The broader question for the robotics and automation space is whether this is a template or an outlier. Textile recycling is a particularly favorable use case for automation: the input material is heterogeneous, the sorting criteria are chemically measurable, and the economic incentive — reducing the 50-percent unrecyclable rate that plagues the industry — is concrete. DataBeyond has framed the Zhangjiagang deployment as a proof of concept for replication elsewhere.
If that replication happens, the number to watch is not the throughput or the accuracy. It is the four workers per shift. Whether reskilling at that scale produces genuine career pathways or just quieter jobs on a louder machine is the question the automation debate has mostly stopped asking. The Fastsort-Textile, at least in Zhangjiagang, has not answered it. It has just made it the right question.
DataBeyond was founded in September 2018. The Fastsort-Textile is deployed at Shanhesheng Environmental Technology Ltd. in Zhangjiagang, installed in 2025. Performance claims in this article are based on company-provided data and have not been independently verified.
† Add † footnote: "Source-reported; not independently verified." Alternatively, attribute explicitly: "the company says the belt moves at 4 meters per second."