The money did not leave fashion technology. It left a genre — avatars, virtual storefronts, digital-only drops — that borrowed the name and never touched a garment. Our position: the retreat investors call a verdict on the category is a correction against theatre, and software that changes what a factory actually cuts is being repriced for someone else's sins. The fix for a founder re-pitching now is not a better narrative. It is refusing the category.
Key takeaways
- The retreat rejects spectacle, not tooling that makes garments cheaper and less wasteful to develop.
- "Fashion tech" is useless as an underwriting unit — the unit is the constraint your product removes.
- A pitch that cannot name the file a factory acts on is still selling pictures.
- We would change our mind if geometry-native tools failed to cut sample rounds in production.
Why did investors stop taking fashion tech meetings?
Ask founders why the meetings dried up and you get a story about sentiment. Sentiment is downstream of something mechanical: for a stretch of the last cycle, the category's most visible deals were spectacle — wardrobes for characters, garments that existed only as renders. Those bets were underwritten on engagement that never arrived, and a committee that hears "fashion tech" now prices that memory.
The supply of capital is not the constraint. Crunchbase counted 195 companies joining its unicorn board in the first half of 2026, above the 193 minted across all of 2025, with robotics and AI leading by count. Money is moving fast, toward companies whose output can be measured in something other than time on a platform.
Venture is also arguing with itself about discipline. In Thrive Capital's first investor letter, reported by TechCrunch in August 2026, Joshua Kushner called the magnitude of the AI opportunity difficult to overstate, then warned that "it would also be a grave error in our minds to let excitement weaken our investment discipline." Hold that against the avatar years: the discipline failure was accepting a rendered garment as a garment. It belongs to a thesis, not to an industry.
What separates real fashion tech from the hype cycle?
One test, and it is not philosophical: does the output enter a system that binds? A pattern file a cutter loads. A marker that decides how much fabric leaves the roll. A tech pack a factory quotes from. Software that makes a picture ends on a screen; software that makes geometry ends as an object with a cost per unit attached.
That test disqualifies much of what is sold as production AI. An image model with a fabric vocabulary is still an image model: it does not know that a bias-cut panel behaves differently from a straight-grain one, and it cannot tell you what happens two grades up. We have argued that virtual try-on cannot show drape, and mapped where fabric simulation's physics ends and the approximation begins. Fabric behaviour is a physical constraint, and a system that has not modelled it is guessing fluently.
The other half of the test is the buyer. In our reading, the sizing and try-on companies of that era mostly ended up inside retailers' and platforms' own stacks rather than as standalone businesses — a sensible home for the technology, an awkward one for a venture return.
What should founders re-pitching in 2026 put in the deck?
Stop selling a category and sell displacement. The pitches that clear a committee now answer these questions before the demo starts.
- The baseline you replace, in the buyer's units: sample rounds per style, patternmaker hours per grade set, fabric per garment, weeks to an approved sample.
- The file you emit, and whether a factory accepts it without redrawing. Interoperability with the CAD and PLM already installed is not a feature; it decides whether you exist.
- Where your training data comes from and why a rival cannot buy the same thing — the argument behind our case that proprietary data is the moat and a model wrapper is not a startup.
- What breaks: tolerances at the size extremes, fabrics that misbehave, factories that reject non-standard output. Naming them makes you easier to underwrite, not weaker.
Executive appetite was never the missing piece. Business of Fashion's account of generative AI reshaping fashion creativity reported that 73 percent of fashion executives said generative AI would be a priority for their businesses, next to the concern that it be adopted without undermining the designer's role. Demand for the outcome exists. What buyers rejected was a rendering pipeline wearing their logo.
What is the strongest argument that the pullback is correct?
The honest counter has nothing to do with avatars. It runs: fashion is a bad software market. Margins are thin outside luxury, IT budgets are small, decisions move at the pace of a season, and the user does not control the budget. Capital is abundant and willing to fund anything with visible pull, so apparel's absence from the winners' column reads as a demand signal rather than a labelling error. The sectors minting new billion-dollar companies are robotics, AI infrastructure, finance, energy and semiconductors — not apparel production.
We concede most of that. A tool that makes a brand rebuild its development process before it returns value loses to the status quo, and plenty of production AI is a rendering pipeline in a trench coat. Deployments here die in integration more often than in the demo.
Where the counter fails is in reading buying behaviour as evidence about technology. Brands never stopped paying for pattern CAD, nesting or PLM; they stopped paying for spectacle. That budget line is smaller than a consumer-app market slide, but it recurs, and it is defended by people whose bonus depends on hitting a sample date.
What would change our mind is production evidence, not pilots: brands running geometry-native tools with no drop in sample rounds, no drop in fabric per garment, no shorter grading cycle, patternmakers still redrawing the output before it reaches the floor. Then the pullback would be right, and we would be wrong about the category rather than only its theatre.
FAQ
Is the fashion tech funding pullback permanent?
Unlikely in the form people describe. Capital is flowing into sectors that show operational results, and apparel development tooling passes that test. What looks permanently gone is underwriting spectacle as though it were infrastructure.
What counts as real fashion tech?
Anything whose output enters a system that binds a physical outcome: a pattern file a cutter loads, a marker that sets fabric consumption, a tech pack a factory quotes from. If the output ends on a screen, it is media, not production technology.
Why did metaverse fashion projects lose investor confidence?
They were underwritten on consumer engagement that never arrived and produced nothing a factory could act on. The technology was not the problem; the thesis was. Investors now discount anything that sounds adjacent, which is why naming matters.
How should a fashion tech founder pitch now?
Lead with displacement, not category. Name the process you replace, the units the buyer already measures, the file formats you emit, the data nobody can copy you on, and the failure modes. Category language invites the memory you are trying to escape.
Further reading
- 40 Companies Joined The Unicorn Board In July, The Highest Count In 4 Years — which sectors are producing new billion-dollar valuations, and which are not.