What Happened to the Sewbot Factory? A Ten-Year Follow-Up

If you have read anything about automation and garment workers in the last decade, you have read about this factory. In 2017 a Chinese apparel manufacturer announced that it would open a plant in Little Rock, Arkansas, where American-built robots would sew t-shirts more cheaply than workers anywhere on earth. The chairman of Tianyuan Garments, Tang Xinhong, put numbers on it in China Daily, quoted widely at the time: "We will install 21 production lines. When fully operational, the system will make one T-shirt every 22 seconds. We will produce 800,000 T-shirts a day for Adidas" (Core77). He added that personnel cost per shirt would be around 33 cents, and that "around the world, even the cheapest labor market can't compete with us."

That quote travelled. It appeared in conference slides, NGO reports, business-school cases and a great many articles about the end of garment employment in Asia. I run a CMT sewing factory in Nepal, so I have had it read back to me more than once by people asking, politely, how long I expected to stay in business.

What I could never find was the follow-up. The announcement was covered by everyone. The outcome was covered by nobody. So I went and read the public record, and this article is what it says.

What was announced in 2017

The technology came from SoftWear Automation, a Georgia Tech spin-out based in Atlanta, whose Sewbots use high-speed machine vision to track individual threads in the fabric and micromanipulators to steer cloth through a conventional sewing machine. IEEE Spectrum's feature described the approach in detail and reported that at the Arkansas factory the robots "will equip 21 production lines, designed to make 23 million T-shirts per year for Adidas" (IEEE Spectrum).

The Robot Report, covering the announcement in September 2017, described a 20 million dollar factory using 330 robots from SoftWear Automation, producing shirts at 33 cents each, with a Sewbot able to make a single shirt in 26 seconds (The Robot Report). The numbers vary a little between tellings, which is normal for an announcement, but the shape is consistent: a fully automated t-shirt plant, at very large volume, for a household-name brand.

The plant was real. It was built, and it opened. In May 2019 the Arkansas Economic Development Commission published a release on Governor Asa Hutchinson touring the newly opened facility, where the company announced an additional 10 million dollars of investment on top of the original 20 million. That release is worth reading closely for one reason: it states that the company "employs 160 and is seeking more applicants," and that it "is on track to hire 400 full-time workers." It also describes Tianyuan as "currently partnering in the development of the first fully automated t-shirt production line" (Arkansas EDC).

Note the tense. Two years after the announcement, at the opening of the plant, the fully automated line was still in development, and the factory was hiring people.

What the factory says about itself today

TY Garments USA's about page is the clearest source available, because it is the company describing its own operation to its own customers. It reads:

"TY Garments USA specializes in manufacturing military and tactical apparel, partnering with renowned brands like Adidas and Reebok. Our 100,000 square-foot facility in Little Rock, Arkansas, is a cornerstone of our operations, featuring 8 advanced production lines designed to meet the highest standards of quality and efficiency. With a $20 million investment and a workforce of up to 200 skilled sewers, our state-of-the-art plant is equipped with the latest technology, including automatic transportation systems and Intelligent Warehousing. Currently, we are also pioneering the development of the first fully automated hoodie production line. Our facility is poised to produce over 4 million garments annually." (TY Garments USA)

Set that against the announcement and the differences are plain. Twenty-one lines became eight. The automation described is material handling and warehousing, not sewing. The phrase "a workforce of up to 200 skilled sewers" is doing the work that 800,000 shirts a day was supposed to do. Neither Sewbots nor SoftWear Automation appear anywhere on the page. And the fully automated line is still described as under development, seven years after the 2019 release used almost the same phrase about t-shirts, now applied to hoodies.

Here is the part I want to be exact about, because it is where most commentary on this story would go wrong. The factory did not fail. It was built, it opened, it operates, it makes over four million garments a year for real brands, and it employs people in Arkansas. By the ordinary standards of manufacturing, that is a success. What did not happen is the robot story. The plant became a garment factory, which is a perfectly good thing to be.

What is not on the record

Neither Tianyuan nor SoftWear Automation has published an explanation of why the deployment does not match the announcement, and I have not found one in the trade press either. So I am not going to supply a reason. I am reading public web pages, not talking to anyone inside either company, and the gap between those two things is exactly where confident-sounding analysis usually goes wrong.

What I can report is what the people building the technology have said about the difficulty. In 2021 SoftWear's chairman and CEO, Palaniswamy Rajan, described the core obstacle as getting to a production system that can "operate 24/7 at high speeds and greater than 98 percent quality," with the added problem that no two batches of cotton behave alike, so fabric and dye variation force recalibration (reported in Mind Matters, citing Wired). That is a statement about how hard the problem is. It is not a statement about this factory, and I am not presenting it as one.

Two other things belong on the record, because the story is usually told wrong in both directions.

SoftWear Automation is alive and funded. In August 2025 it closed a 20 million dollar Series B1 round led by BESTSELLER, the Danish fashion group behind JACK & JONES, VERO MODA and ONLY, invested through BESTSELLER's Invest FWD platform, with BESTSELLER's CFO Thomas Børglum Jensen joining the board (PR Newswire, 11 August 2025). A good deal of secondary coverage dates this round to February 2026; that date is wrong, and if you are citing it, cite the release. The detail I find most telling is who wrote the cheque. Nine years after Little Rock, the buyer of automated sewing is a brand looking for production closer to its customers, not a contract manufacturer in Asia chasing unit cost.

Sewbots work — inside their envelope. IEEE Spectrum reported in 2017 that SoftWear's machines were already making bath rugs, pillowcases and towels, with two million such products on sale at major retailers. Flat, simple, mostly square or round. Rajan drew the boundary himself in the same piece: "High fashion, bridal dresses, things like that — those are still going to be done by humans." Nothing in the ten years since contradicts that. The honest claim is not that the robots do not work. It is that the envelope did not widen as fast as the press releases implied.

Why seam-joining resists automation

The reason this particular problem keeps defeating well-funded, technically excellent teams is worth understanding, because it is the same reason on every floor.

A robot arm is good at rigid parts. A rigid part is where you put it, and it is the same shape on the tenth pick-up as on the first. Fabric is not like that. Two plies of jersey shift, stretch and curl differently every single time they are picked up, so the gripper has no stable reference and the vision system has no repeatable target. The material deforms under the very act of being handled, which means the machine's model of where the cloth is becomes wrong the moment it touches it. SoftWear's response to this was genuinely clever — instead of tracking the panel's outline, track the individual threads with a camera capturing over a thousand frames a second, and correct continuously. That is a real solution to a real problem, and it is why the company got as far as it did.

But on my floor the variation goes further than the physics of a single seam. Fabric arrives in shade lots that behave differently. Knits relax by different amounts depending on how long they sat after cutting and how humid the room is. An interlining that was fine last month feels different this month. A sleeve in size S and the same sleeve in size XXL do not present the same way to a feed dog. My operators absorb all of this without anyone writing it down: they feel a fabric that is running tight and ease it, notice a ply that has curled and flip it, and they do this a few hundred times a shift without being asked. That absorption of variance is the job. It is not a residual task left over after the sewing.

This is also why the 98 per cent figure matters more than it looks. A line running at 98 per cent quality at t-shirt volumes is generating a very large number of defective pieces per day, and on a fixed automated line there is no operator to catch the drift at the station where it starts. My checkers find problems at the table, but a good operator finds them at the machine, three pieces in, and stops. Automation that removes the operator also removes the sensor.

What this means if you own a CMT factory

The practical question is whether to hold back on investment because a machine is coming that will change the economics. Ten years of evidence says no, and this factory is the cleanest test case available: the single most heavily promoted robotic sewing deployment in the industry's history, with real money, real brand demand, a nearshoring advantage and competent engineering behind it, now runs with up to 200 skilled sewers.

That does not mean nothing is happening. Work on robotic apparel assembly continues, and some of it is serious — the ARM Institute's robotic sewing project with Siemens, Sewbo, Levi Strauss and Saitex is publicly documented and has targeted real operations on denim (ARM Institute), and CreateMe has pursued adhesive bonding as a route around the needle entirely (CreateMe). These are worth watching. None of them is standard equipment in a CMT factory, and none is likely to be in your next budget cycle.

Meanwhile the automation that genuinely changed costs in the last decade was unglamorous and is already on the market. Spreading and cutting went CNC, which is mechanisation rather than AI, and it is where measurable labour change has actually shown up — a point I take up separately in the piece on the 88 per cent statistic and what the ten-year scoreboard shows. Machine heads got sensors and telemetry. And the cheapest thing available to most factories is still a record of what happened: which operator did which operation on which bundle, on which machine, at what time. I have written at more length about what AI actually runs on a sewing floor in 2026, and the short version is that the factories which will benefit when any of this matures are the ones that spent the intervening years recording their own production. There is a broader survey of what pays back in our guide to garment factory automation in 2026.

If you are planning capacity for 2027, plan for people sewing your garments, and spend the money on the things that already work.

The limits of this analysis

I should be clear about what this article is and is not. I have read company web pages, a state economic development release, contemporaneous trade and technical journalism, and a funding announcement. I have not spoken to Tianyuan, to TY Garments USA, or to SoftWear Automation. No party has confirmed or denied any characterisation here, and if any of them publishes a fuller account I will update this page.

A company's about page is a marketing document, not an audited disclosure. The absence of Sewbots from it is a real and checkable fact about what the company chooses to say it does today, but absence from a web page is not proof that no Sewbot exists anywhere in the building. Equally, announcement figures quoted in 2017 were forward-looking statements by an executive, not commitments. My claim is narrow and I will restate it plainly: the configuration that was announced is not the configuration the company describes today, the difference is large, and no public explanation exists.

That is a smaller claim than the headlines this story generated in 2017. It is also, I think, the only one the evidence supports.

Frequently Asked Questions

Did the Sewbot factory in Arkansas fail?

No. The plant was built, it opened, and it operates today. TY Garments USA describes a 100,000 square-foot facility in Little Rock producing over 4 million garments a year for brands including Adidas and Reebok. What did not happen at the announced scale was the robotic part. The 2017 announcement described 21 Sewbot production lines making 800,000 t-shirts a day; the company's own about page today describes 8 production lines and a workforce of up to 200 skilled sewers, and does not mention Sewbots or SoftWear Automation at all.

Why did the Sewbot deployment not reach the announced scale?

Neither company has published an explanation, and anyone who tells you the reason with confidence is guessing. What is on the record is that SoftWear Automation's own CEO described the hard part in 2021 as operating 24 hours a day at high speed above 98 per cent quality, with no two batches of cotton alike. That is a description of a difficulty, not a stated cause. The honest summary is that the announced configuration is not what the plant runs today, and the public record does not say why.

Is SoftWear Automation still in business?

Yes. The company is active and funded. In August 2025 it announced a 20 million dollar Series B1 round led by BESTSELLER, the Danish fashion group behind JACK & JONES, VERO MODA and ONLY, made through BESTSELLER's Invest FWD platform, with BESTSELLER's CFO joining the board. Some coverage dates this round to February 2026, which is wrong; the release is dated 11 August 2025. Note who bought in: a brand, not a contract manufacturer.

Can robots sew a t-shirt today?

Within a narrow envelope, yes, and Sewbots genuinely work inside it. The published scope has been flat, simple, high-volume goods: bath rugs, pillowcases and towels were the products IEEE Spectrum reported as already shipping in 2017, and the company's CEO drew the limit himself with the line that high fashion and bridal dresses will still be done by humans. A general sewing line producing mixed styles, sizes and fabrics is a different problem, and it is not solved. If you are budgeting for 2027, budget for people sewing your garments.

What should a CMT factory owner do about sewing automation?

Do not plan capacity around a robot arriving. Ten years after the most heavily promoted deployment in the industry's history, the flagship plant runs with sewers. The things worth buying in that window are the ones that already work and pay back: cutting-room automation, better machine heads, and a record of what your floor actually did. Waiting for the robot costs you the decade in which the unglamorous investments would have compounded.

The lesson I take from Little Rock is not that automation is a bluff. It is that the industry has a habit of reporting announcements as outcomes, and then never going back. Nine hundred words of coverage in 2017; nothing since. If you are making a ten-year capital decision on the strength of a press release, go and read the company's about page first. It took me four minutes, and it said more than the decade of commentary built on top of it.

Santosh Rijal is the founder of Scan ERP, a garment manufacturing ERP system designed for factory floor operations. He works directly with sewing lines, cutting rooms, and production supervisors across Nepal's garment manufacturing sector.