3D Knitting & On-Demand Manufacturing

A WHOLEGARMENT knitting machine takes in one strand of yarn and emits a finished sweater. No cutting. No sewing. The only scrap is the tail at the end of the spool. Cut-and-sew manufacturing, by contrast, throws away 15–25% of every bolt of fabric as off-cuts on the factory floor — roughly 92 million tonnes globally per year, before a single garment is worn. 3D knitting does not improve that number. It abolishes the category.

How it works

Conventional garment-making weaves flat fabric in bulk, prints a 2D pattern on it, cuts the pieces out, and sews them together. Each step adds waste and labor; the seams it leaves behind become the first places the garment fails.

A WHOLEGARMENT machine — Shima Seiki's term for the technique it commercialized in 1995 — works differently. The garment exists first as a digital file describing 3D shape plus yarn behavior. The machine then knits the entire object as one continuous loop, stitch by stitch. Every loop is independently controllable, so density, texture, and 3D curvature can vary across a single piece: a dense sole grading into mesh upper for a shoe, a curved heel cup in a sock, ribbing at a cuff that thickens into a cable up the arm.

The 2025 SES-R adds a spring-type moveable sinker, which lets the machine shape parts of the garment out-of-plane mid-knit — the closest a flatbed knitter has come to sculpting in three dimensions while the yarn is still in motion.

Specific examples

Nike Flyknit (2012 launch): the running shoe upper as a single knitted sock. Saved Nike a reported 3.5 million pounds of manufacturing waste in its first decade and diverted 4+ billion plastic bottles into recycled polyester yarn. The 2024 core Flyknit yarn is 50–85% recycled polyester.
Adidas Knit for You (Berlin pop-up, 2017): 3D body scan in-store, custom merino sweater off the machine in roughly four hours. The first industrial demonstration of a bespoke knit garment manufactured at point of sale.
Medical compression garments: per-patient WHOLEGARMENT production is now commercial for compression hosiery and orthopedic supports — applications where individual fit matters more than unit cost.
Shima Seiki SWG-XR: four-needle-bed machine using SlideNeedle to do all-needle knitting at fine gauges. Its companion software, APEXFiz, in April 2026 added a plugin that pipes garments straight from CLO's 3D fashion design tool into machine production files, removing the manual translation step that used to take days.

A complex WHOLEGARMENT item still takes 20–120 minutes to come off the machine. That is roughly an order of magnitude slower per unit than cut-and-sew at commodity scale, and roughly an order of magnitude faster than any other route to a custom-fit garment.

The on-demand shift

Fashion runs on forecasts: design 6–18 months ahead, manufacture in bulk against a guess, hold inventory, discount whatever didn't sell. Industry estimates put the forecast error around 30–40% of volume. That gap is overproduction, and most of it ends up incinerated, landfilled, or shipped to secondary markets where it again fails to find a buyer.

A machine that knits one garment per order replaces the forecast with a response. The consequences cascade: no minimum order quantities, no seasonal collections, no end-of-season markdowns, no warehouse of unsold inventory. A machine in a retail back-room can plausibly replace a factory in another time zone. The 3D knitting machine market sits at $1.4B in 2025; the forecast for 2035 is $2.4B (CAGR 5.6%) — small relative to global apparel, but the leverage point is not the machine count, it's where the machines live.

What's contested

Three places the optimism breaks down.

Speed vs. scale. Cut-and-sew remains drastically faster per unit at the volumes that move the apparel market. No path yet exists to make WHOLEGARMENT competitive for a t-shirt that sells for $8. Multi-head and parallel-needle-bed designs are the active R&D frontier, but no commercial machine has closed the gap.

Yarn compatibility. The technique works best with continuous-filament synthetic yarn. Cotton, wool, and linen are shorter-staple, fuzzier, and more likely to snap under machine tension. So the cleanest-manufacturing technique currently couples most easily to the least biodegradable input. Hybrid blends paper over the conflict but make end-of-life recycling harder.

The seam-free recycling promise. A mono-material WHOLEGARMENT garment should be straightforward to chemically recycle — one polymer, no thread, no interlining. In practice, performance demands push designers toward blended yarns, which break that recyclability story. Whether the industry converges on mono-material yarns or on compatible multi-material recycling is unresolved.

Why this has to do with other realms

The Shima Seiki machine is a direct lineal descendant of the tech jacquard loom (1804), the first machine in history to act on encoded external data via punched cards. The Jacquard's instruction-card lineage runs Jacquard → Hollerith tabulator → von Neumann architecture → modern computer. Less famously, it also runs Jacquard → flatbed knitter → WHOLEGARMENT → present-day digital fabrication. Same idea, two branches: a machine that reads its instructions instead of performing a fixed operation. The textile industry was doing software-controlled manufacturing 140 years before software had a name.

There's a quieter biological parallel. A WHOLEGARMENT machine produces a 3D shape from a linear sequence of encoded stitch instructions, with local stitch chemistry determining global form. Proteins do the same trick: a linear amino-acid sequence folds into a 3D shape because of local chemistry between adjacent residues. The myelin sheath around a neural axon thickens or thins in a position-coded pattern to tune signal speed — the biological version of stitch-density grading. Whether this is structural analogy or something deeper is a question the biomimicry literature has barely touched.

An open question

If a knitting machine can sit in a retail store and produce a bespoke garment in four hours, what is the right unit of fashion design — the garment, or the program that generates it? And who owns the program?

Key sources

Knitting Technology by David J. Spencer (Woodhead, 3rd ed. 2001) — the standard reference on flatbed and WHOLEGARMENT mechanics; load-bearing for the machine description.
Shima Seiki technical documentation on WHOLEGARMENT, SWG-XR, SES-R, and APEXFiz (manufacturer site) — primary source for machine capability claims.
Ellen MacArthur Foundation, A New Textiles Economy (2017) — origin of the 92Mt/year and overproduction figures that frame the on-demand argument.
To verify: independent peer-reviewed LCA comparing WHOLEGARMENT to cut-and-sew on cradle-to-gate waste and energy. Manufacturer-cited 60–90% waste reduction is plausible but not independently audited at scale.
Nike sustainability reporting (annual Impact Report) — source for Flyknit recycled-bottle and waste-diversion figures.

Abhishek's take

The 30-40% forecast error the body cites is the number I live inside every season: I book styles four to six months out on a guess, and if the guess is wrong, the markdown is the confession. Where 3D knitting becomes interesting to a buyer isn't replacing cut-and-sew at volume; the speed gap is too large for that. It's in test quantities — knit four color variants against confirmed early-season signal before I commit the full minimum order to a vendor six time zones away. That's a different bet, and it's one I'd take.

Abhishek