AMPulse

Founding cohorts 2012-2023

The Aiming of a Generation

Booms do not kill their vintage. They aim it. Where an AM company aimed at founding predicts its odds of dying far better than the year it was born, and the destination that looks safest did not keep its companies either. It sold them.

AMPULSE DATA REPORT · 2026Data snapshot: Sep 7, 2026

Where AM companies die

The 2012-14 founding window is remembered as the deadliest in additive manufacturing, the desktop boom's generation. The belief that follows is that the year a company was born set its odds. Counted whole, the window says something else. Of the 173 AM companies founded between 2012 and 2014 to sell to consumers, 56 are recorded as defunct. Of the 62 founded in the same three years to sell to aerospace and defense, 1 is.

Same founding years, same decade of exposure, same shocks. Inside one cohort the recorded odds of dying run from 32.4% down to 1.6%, and the thing separating the two ends is where the company was aimed.

The middle of the chart is where the risk intuition breaks. Industrial is the cohort's largest destination by far, 685 companies, and it sits at 10.4%, below Education. Medical sits beside it. The destinations that demand capital equipment, qualification cycles and years of unpaid engineering are the survivable ones. The destination a founder could enter from a garage is the one that killed the largest share of its entrants.

Death is one of two ways to leave a cohort. Add the companies that were bought and the picture of the safest destination changes.

10 of the 62 aerospace and defense companies were acquired, 16.1%, the highest acquisition rate of any destination in the cohort and more than double Industrial's 7%. Count both terminal routes together and the destination that looks safest on the mortality chart is unremarkable: 17.7% of the cell stopped being an independent company, against 15.6% for Education, the destination the first chart calls second-deadliest. Aerospace and defense companies founded in this window did not die. They got bought, and the fourth chapter is about who bought them and why that is the finding.

The unit of death was the desktop printer

A destination is a coarse label, and the mortality inside each one is not evenly spread. Cut the same cohort by destination and value-chain position together and the spread becomes a ranking of business shapes.

The top of the ranking is one cell. 30 of the 62 companies founded in 2012-14 to sell printers to consumers are gone, 48.4%. The consumer services around them, print bureaus and marketplaces, died at 24.4%; consumer materials, mostly filament sellers, at 14.8%. The machine was the thing that failed. The names are the ones a reader of that era will recognize:

3DPandoras, 3DSTUFFSNL, Active3D, AiO Robotics, Alligator Board, Bad Devices, Buildatron, Choc Edge, CKAB, Cohesion3D, DreamBox 3D, Dynamo3D, Electroloom, electronic things, iBox Printers, Ira3d, LONGER, M3D, MakerFarm, OpenBeam, Peachy Printer, Pirate3D, Replikat, RigidBot, Sculptify, Stalactite 3D, UltiBots, XYZprinting, Zeepro, Zeus.

Inside Industrial the same line runs through the technology column rather than the value chain. Of the 435 industrial-aimed polymer companies in the cohort, 13.1% are defunct; of the 95 metal ones, 2.1%. Ceramics and composites lost none of 24. A polymer extrusion company aimed at industrial customers in 2013 shared its cost structure and its competitors with the consumer printer makers above, and shared their fate at a lower rate.

Why the desktop machine could not hold a company is not something a census proves, and this report does not try to. The pattern the survivors show is the interpretation AMPulse works from: the hardware commoditized within a few years of each wave, and the companies still standing in that segment are the ones whose moat moved into firmware, cloud slicing and a model repository, the platform layer rather than the frame and the hotend. The 2012-14 cohort mostly shipped frames.

The boom that aimed a generation

Cohort-wide, the 2012-14 window looks the way its reputation says it should: 154 of 1,107 companies, 13.9%, are gone.

Most of that came from one aim. Consumer and Education together took 269 of the 1,107, 24.3% of the generation, and produced 69 of its 154 deaths, 44.8%. Their combined mortality is 25.7% against 13.9% cohort-wide. A quarter of the generation supplied nearly half its deaths.

The industrial side gives a different answer depending on which axis you ask. Hold the three industrial destinations fixed, Industrial, Medical and Aerospace/Defense, and compare the window against the next one: 81 of 828, 9.8%, against 58 of 897, 6.5%. On destination alone the later companies did better, which is what a vintage story predicts.

Now narrow the same three markets to metal and ceramics. 2012-14: 3 of 170, 1.8%. 2015-17: 14 of 271, 5.2%. The comparison reverses, and the 2015-17 cell had three fewer years in which to lose members and still lost the larger share, so the age gap runs against the finding rather than producing it.

What the 2012-14 window did to its generation was not expose it to a bad year. It sent a quarter of it at the destinations that commoditized, and split the rest along a line the destination label does not record.

Defense did not die. It was absorbed

The aerospace and defense cell of 2012-14 is 62 companies, 59.7% of them American, 24 holding an active patent link and 38 with a recorded funding round. Who was acquired inside it is not random.

The 18 machine builders lost nobody and sold nobody. The acquisitions sit in the two positions closest to the customer's part: 4 of 19 service bureaus and 4 of 14 companies that used additive to make their own aerospace product. Where the record names the buyer, the buyer is a prime or a larger operator in the same supply chain: Terran Orbital went to Lockheed Martin, UTC Aerospace Systems into Collins Aerospace, Deep Space Industries to Bradford Space, 3D Material Technologies to Aerojet Rocketdyne, Identify3D to Materialise. The database records a parent for 1 of the 10; the rest of those names come from the public record of each deal.

This is the mechanism a decade of aerospace additive has taught, and the cohort is its statistical trace. Qualifying a printed part for flight takes years and the qualification belongs to a program, not to a technology. Once a bureau or a parts maker holds that position on somebody's part number, it is worth more inside the prime than on the merchant market, and the qualification transfers with the company or not at all. So the safest destination does not keep its suppliers independent. It absorbs them, and the merchant supply base for defense additive shrinks by acquisition in the same years the consumer base shrank by closure. One of those shrinkages makes headlines. The other makes a line in a 10-K.

The cell's recorded zero is now a one, and the correction has more in it. 4 of its 51 live-status rows carried a liveness flag at the snapshot and each was re-checked by hand: AMC Powders Co., Ltd. is gone, its domain surrendered to a host-suspension notice; International FemtoScience, Inc. and Gilmour Space are alive and the probe misread them; PartnersLab no longer resolves and could not be settled either way. The defensible mortality of the cell is therefore 2 to 3 of 62, 3.2% to 4.8%, against 32.4% two bars away.

Two instruments, no agreement

Fusion3 passed a routine AMPulse verification on 2026-07-21 while it was in liquidation. Trade coverage of the liquidation and the asset auction ran through July; our own audit caught the row and reclassified it on 2026-08-24. Dead domains rarely 404 any more. They answer HTTP 200 from a parking page, a domain-sale listing, or a redirect to an unrelated host, which is exactly what a working company looks like to anything reading a status code.

The liveness probe this report rests on was rebuilt after its first run, when a bug turned out to have discarded 622 network failures as noise, including every domain that had stopped resolving. With that fixed, the signal that fires most often is the least ambiguous one: at the snapshot, 207 of the 513 live-status rows carrying a flag have no working DNS, more than any other signal, and 270 in all carry a signal that means the web presence is gone rather than moved or thin. A public-resolver check of the first 232 DNS failures found 166 with no address and no nameserver at all.

A flag is still not a death, so forty flagged rows were adjudicated by hand against every independent trace that could be found. 16 of the 40 no longer exist as listed, 12 abandoned and 4 absorbed by acquisition; 11 had moved domains and were fine; 13 were the probe's mistake. Of the 16 exits, 11 had no public record anywhere: no filing, no trade story, no farewell post. Every acquisition was announced. Nearly every closure was not.

That ratio is the reason recorded mortality cannot be compared across countries, and the country table is where it shows.

The two series rank the same countries almost independently of each other; the rank correlation across the 14 is -0.2. Britain records the highest mortality of the 14, 17.1%, and one of the lower domain-loss rates; China records 1.1% mortality and loses domains at 5.5%, 5 times the rate; Korea and India are the same shape. Germany and France sit at the other end, with deaths recorded and domains kept. Neither series is mortality. Recorded mortality measures how loudly a country's failures are announced: in Western Europe a wind-up produces a filing and a trade story, often while the domain stays paid up. Domain loss measures whether anyone kept paying the registrar. Each carries the announcement culture of the country it counts.

For this report the consequence is specific. No cross-country mortality claim appears in it, and no regional control is offered for the destination finding, because the instrument that would supply one measures announcement rather than survival. The destination comparison survives because the undercount runs the same way the record does: among the still-live rows of the 2012-14 cohort, Consumer carries a flag on 17 of 106, 16%, and Aerospace/Defense on 4 of 51, 7.8%. Credit every flag as a death and the two ends of the first chart move apart, not together.

The re-aiming

Two destinations from the first chart have swapped roles in the founding data. Education, second-deadliest in the 2012-14 cohort, has nearly stopped receiving new AM companies: 8.7% of that window's formations, 2% of the next, 0.8% of 2020-23. Aerospace and defense went the other way over the same three windows, 5.6% to 10% to 14.3%.

Defense holds the best mortality record in this data, and it was measured on a kind of company that the new cohort mostly is not. The 2012-14 cell was 25 machine and materials suppliers out of 62, selling into a budget insulated from the demand cycle, and its survivors were absorbed by the primes they supplied. The 2020-23 cell is 123 companies, 51.2% American and 17.9% Chinese, and its value-chain labels say little about it: the census files Vast and Impulse Space as hardware makers. Its funding says more.

Eleven of the twelve sell no printer, powder or slicer to anyone; the twelfth, Redwire, builds in-space manufacturing systems for government customers. They build drones, spacecraft, hypersonic weapons and launch vehicles, and they are in an additive census because additive is inside their production. 8 of the cohort's 123 companies have already taken a round of $100M or more, within five years of founding. The industry is not sending more printer makers at defense budgets. It is sending defense product companies that print, and their survival will be decided by whether a program converts, the gap between an award and a production line that the last cycle of defense additive left mostly open. The destination's record cannot be carried forward to them, because it was earned by suppliers to a budget and they are bets on the budget itself.

The capital arriving at the cohort changed shape at the same time. The median disclosed AM round moved from $2.2M in 2024-Q1 to $16.6M in 2026-Q2, with the middle half of rounds widening from $0.5M to $10.5M into $3.3M to $71.3M, and disclosed rounds under $2M falling from 73 a quarter to 17. The record behind that number is narrow: AM companies only, disclosed amounts only, and 1,010 of its 1,030 events arriving through one structured channel.

One month of that record, July 2026, from trade coverage AMPulse tracked and the companies' own statements. 3DEO filed for insolvency, and its entire IP estate and machine fleet opened at $3.43M against roughly $80M raised. BigRep entered liquidation and delisted from Frankfurt. Fusion3 liquidated and its assets went to auction. Wurth Additive Group ceased operations. In the same month Beehive Industries ordered 30 EOS M4 ONYX systems, over $50M by the company's account, alongside a $70M Ohio expansion; VulcanForms opened a third Massachusetts facility; Velo3D took a 289,000 square foot campus for Forge 1. The small end is closing and the large end is buying machines, and the buyers are the companies with a part number.

The counter-thesis

The strongest objection is that the destination is our label, applied once per company at the snapshot, and a company that started in consumer and survived by moving to industrial is counted as an industrial survivor. If pivots ran that way, Consumer's mortality is inflated by the departures of its winners. Two things bound it. The technology-based cell that does not depend on the market label, every consumer or education company plus every polymer hardware maker regardless of market, dies at 21.5% against the label-based 25.7%, so the label is not manufacturing the spread on its own. And the dead are named above; a reader who knew that market will not find a mislabelled industrial company among them.

The second objection is exposure. A cohort that has had twelve years to fail cannot be compared with one that has had four, which is why every comparison in this report is made inside one founding window or between the same cell in two adjacent windows, and why the 2020-23 defense cohort's mortality is never quoted.

The third is that the census entered late. A 2013 desktop startup that folded in 2015 left less indexable trace than a metal company that folded the same year, so detection correlates with capital intensity and some of the first chart's spread could be the index. That objection is conceded and it cuts in one direction: every mortality figure here is a floor, and the floor is lowest where the trace was thinnest, which is the consumer end.

The fourth is the one the fifth chapter makes against the report itself. Recorded mortality is announcement culture. It is the reason the report makes no claim across regions and confines itself to comparisons where the undercount was measured and runs with the finding.

What to watch

• The 2020-23 defense cohort at age ten. Run the destination query against it when it reaches the age the 2012-14 companies have now and read where it lands: inside the old cell's 0.3-8.6% interval, or near the 10.4% that Industrial carried through a full cycle. The answer separates a safe destination from a safe kind of company.
• Program conversion in the most-funded twelve. A named production award, not an SBIR or a demonstration contract, for each of them. The first to close without one is the first data point on what the destination's record is worth to a product company.
• The DNS queue. 207 live-status rows with no working domain, draining through source-based adjudication. Every one that closes as a death raises a mortality figure in this report, and the consumer cells will move first.
• Absorption in defense. The next acquisition of a qualified additive supplier by a prime. The 2012-14 cell says this is how the destination resolves its survivors, and the merchant supply base for defense parts gets smaller each time.
• Consumer, second wave. The 2020-23 consumer cell is 12.6% of its cohort. Its hardware mortality at age five, against the 48.4% the 2012-14 hardware cell reached at age twelve, is the first reading on whether the platform layer changed the odds.

Lee, S. (2026). The Aiming of a Generation. AMPulse Data Reports. https://www.ampulse.online/reports/am-destination-mortality-2026