AMPulse

Cohort analysis

The Industrialization of AM

Three concurrent shifts have remade the composition of new AM company formation across the 2010s and 2020s. The 2020-2026 cohort has settled into the new shape.

Value-chain share of new AM companies · by formation cohort

Hardware

Service

Application

AMPULSE DATA REPORT · 2026Data snapshot: May 14, 2026

Setup: three shifts, one direction

The composition of newly-founded additive-manufacturing companies looks different in the 2020s than it did in the 2010s, and the difference is concentrated in three categories: service value-chain position, polymer-AM material focus, and the aerospace-and-defense target market. Service-VCP share of new AM companies fell from 34% in the 2010s to 25% in the 2020s. Polymer-AM share dropped from 56% to 40% as Metal-AM share rose from 19% to 25%. Aerospace/Defense target-market share roughly doubled, from 6.6% to 13.0%, while Education collapsed from 5.7% to 0.9%.

The shifts mostly resolved by 2020. The 2024-26 cohort is too small (n=163) to draw fresh conclusions from on its own, but the 2020-23 cohort (n=766) already sits within a few percentage points of the 2024-26 cohort on most of these dimensions. This report does not claim the shifts are currently accelerating. It claims they happened and the current cohort reflects the result.

Two categories did not shift meaningfully. Hardware as a value-chain position was 34% of 2010s formations and 35% of 2020s formations (within a percentage point). Industrial as a target market was 66% across the 2010s and 63% across the 2020s (within three points and still the dominant target market in both decades). The shifts are in categories around hardware and industrial, not in those two categories themselves. The hero panel above shows the three categories that moved, with sparklines for the four five-year cohorts that span the analysis window; the two categories that stayed flat are deliberately not in that panel.

The full database scope for this analysis is N=2,811 companies founded 2010-2019 (the "2010s" decade aggregate) and N=929 companies founded 2020-2026 (the "2020s" decade aggregate). All numbers in the body sections that follow are decade aggregates unless explicitly noted as cohort-specific; per-cohort detail lives in the charts that anchor each section and in the frozen JSON files linked from the methodology box.

The report runs through three axis sections (one per shifted category), a short sidebar on micro-shifts within the 2024-26 cohort, three audience-tagged implications, and a closing list of the asterisks that matter.

§ 1: Service down, application up

Service VCP share fell from 34% of new AM company formations in the 2010s to 25% in the 2020s. Over the same window, Application VCP share rose from 5% to 11%. Hardware VCP share stayed essentially constant: 34% in the 2010s and 35% in the 2020s, within a percentage point at the decade level. The composition movement within the value chain is therefore Service-to-Application, not Service-to-Hardware. Service fell 9pp; Application rose 5pp. Most of the value-chain reshuffle is accounted for by these two categories; the residual is spread thinly across Materials, Software, Platform, and Post-Processing positions, each of which moved by less than 3pp at the decade level.

The Service decline is monotonic across the four five-year cohorts: 36.8% in 2010-14, 30.9% in 2015-19, 25.2% in 2020-23, 23.9% in 2024-26. Most of the decade-aggregate fall (−9pp) occurred between 2015-19 and 2020-23 (−5.7pp). The 2020-23 → 2024-26 step is a single percentage point and well within the small-sample noise band at n=163. At the cohort level, the Service-bureau era's share of new formations peaked in 2010-14 and has been at roughly its 2020-23 level for the four years since. Service is still the second-largest VCP category in the 2020s, below Hardware and above Application; it did not collapse.

The Application rise is also monotonic: 3.9% / 6.6% / 10.4% / 11.0% across the four cohorts. The largest step is between 2015-19 (6.6%) and 2020-23 (10.4%), a +3.8pp move that did most of the decade work. The Application VCP groups companies whose primary product is something made using AM (an aerospace component, a medical device, a defense platform) rather than the printers, materials, software, or services that the rest of the value chain provides. Spotlights below include three 2020s Application archetypes: Magnet Defense (autonomous unmanned surface vessels), KeratOPrinter (bioprinted human corneas), and Astrobase (reusable launch-vehicle ecosystem). The first two are end-products served to a specific customer. Astrobase is the boundary case in which the company makes both an AM-enabled product and the AM hardware that builds it; the AMPulse classification places it under Application because the public-facing offering is the rocket, not the printer. These are the cases that did not exist as new formations at meaningful share in the 2010s.

Hardware as a value-chain position was the largest category in both decades, and that position is unchanged at the decade level. Three of the eight Group B spotlights below (Deep Manufacturing, GRU Space, Creator Precision) are Hardware-VCP 2020s archetypes representing variations on the Hardware theme: synchronized multi-robot WAAM at industrial-construction scale, lunar-infrastructure printers, and Electric-Field-Driven jet deposition for sub-micron precision. The Hardware category is flat across the decades; the kinds of hardware new companies build, the capital intensity at which they build it, and the target markets they build for are what §§ 2 and 3 describe.

This report describes the share movement and does not assert who absorbed the displaced service-bureau demand. The largest public AM service businesses (Protolabs, Xometry, Materialise) published service-revenue figures in their filings over the same window, and readers interested in following that thread can check those filings directly. The present analysis stops at AMPulse formation data, which is informative about the shape of new companies but silent on incumbent revenue dynamics.

§ 2: the polymer-metal reshuffle

Polymer-AM share of new AM company formations fell from 56% in the 2010s to 40% in the 2020s. Metal-AM share rose from 19% to 25%. Polymer remains the plurality material category in the 2020s; it did not disappear, it ceased to be a majority. The ratio of polymer-AM formations to metal-AM formations moved from roughly 3:1 in the 2010s to roughly 1.6:1 in the 2020s.

The Polymer decline did most of its work early. The cohort sequence is 64.2% (2010-14) → 48.4% (2015-19) → 38.9% (2020-23) → 46.0% (2024-26). The 2010-14 → 2015-19 step alone is −16pp; by 2015-19, polymer had already shed roughly a quarter of its share. The 2020-23 reading (38.9%) was the lowest in the four-cohort sequence; the 2024-26 reading (46.0%) is a +7pp partial rebound that § 4 returns to. At decade-aggregate level, the 2020s as a whole sits at 40%, weighted heavily toward the larger 2020-23 cohort.

The Metal rise also did most of its work early. The cohort sequence is 13.8% / 23.6% / 25.9% / 23.3%. The +10pp move between 2010-14 and 2015-19 is the dominant step; metal-AM share has been within a ±2pp band of 24% across all three subsequent cohorts. The 2024-26 dip of −2.6pp from the 2020-23 peak (25.9 → 23.3) is a single-point move within the small-sample noise band at n=163.

The other material categories together remained a long tail across both decades. Multi-Material moved from 4.9% to 5.2% at decade aggregate (essentially flat). Bioprinting moved from 3.8% to 3.3% (slightly down). Ceramics/Composites moved from 4.5% to 6.2% (modestly up). Of the eight 2020s-cohort spotlights below, Deep Manufacturing and Astrobase are Metal-AM archetypes. GRU Space (lunar-infrastructure printers) and Co-reactive (CO2-mineralizing concrete) are both Construction AM. KeratOPrinter (bioprinted corneas) and Magnet Defense (Multi-Material via hybrid AM) round out the material spread. The Magnet Defense pattern of combining several AM modes for a single end-product is a 2020s composition feature to track, even at single-digit share.

Two of the categories shown above (Construction AM and Novel Process) were redefined in April 2026 by AMPulse taxonomy updates. Construction AM was split out as its own top-level category (previously sub-tagged under Polymer AM or Novel Process), and Novel Process itself was partially redistributed to more precise sub-categories. The visible Construction AM growth (1.7% → 5.6% at decade level) is partly a classifier-evolution artifact and is flagged in the chart caption; this report does not use it as evidence for the thesis. The methodology box has the full disclosure.

This report describes the share movement and does not assert why metal grew or polymer shrank. The why question is not derivable from AMPulse formation data alone.

§ 3: Education collapse, aerospace-and-defense rise

Aerospace/Defense target-market share of new AM company formations roughly doubled, from 6.6% in the 2010s to 13.0% in the 2020s. Over the same window, Education target-market share collapsed from 5.7% to 0.9%. Industrial remained the dominant target market in both decades (66% in the 2010s and 63% in the 2020s), at a magnitude that no other target market approaches. The composition movement within target markets is therefore concentrated on the A/D rise and the Education collapse, against a stable Industrial majority.

The A/D rise is monotonic across the first three cohorts and plateaus thereafter. The sequence is 4.7% (2010-14) → 8.2% (2015-19) → 13.1% (2020-23) → 12.9% (2024-26). The +3.5pp step from 2010-14 to 2015-19 sits entirely inside the 2010s. The largest step (+4.8pp) crosses the decade boundary from 2015-19 to 2020-23, so the headline "2010s → 2020s" framing folds that move into the decade aggregate. The 2020-23 → 2024-26 step is essentially zero: by the 2020-23 cohort, A/D share had reached its decade-level value and the 2024-26 cohort sits at the same level rather than continuing to climb. Spotlights below include three 2020s A/D archetypes: Magnet Defense (autonomous unmanned surface vessels), GRU Space (lunar infrastructure for civil-space and dual-use), and Astrobase (reusable launch-vehicle ecosystem). A detailed deep-dive into the A/D rise (sub-segmentation by drone, missile, satellite, ground-vehicle, naval, and soldier-system applications; geographic concentration; policy environment) is reserved for Report #6 in this series.

The Education collapse is also concentrated early. The sequence is 9.8% (2010-14) → 2.1% (2015-19) → 0.8% (2020-23) → 1.2% (2024-26). The 2010-14 → 2015-19 step alone is −7.7pp; by 2015-19, Education share had already fallen to roughly a quarter of its 2010-14 value. The 2024-26 reading of 1.2% is a +0.4pp uptick from 2020-23 within the small-sample noise band at n=163. The MakerBot-era school/maker target-market position (desktop polymer-AM hardware sold into K-12 and university markets) was effectively a 2010-14 phenomenon. New AM companies targeting Education as the primary market have been at or below 2% share across the three subsequent cohorts.

The other target markets together held shape across the decades. Consumer moved from 11.8% to 12.5% (essentially flat with cohort-level noise: 14.4% / 9.5% / 12.3% / 13.5%). Medical moved from 9.0% to 8.7% (essentially flat). Automotive moved from 0.9% to 1.6%, a small absolute change that roughly doubled its share but stayed below 2% in every cohort. Industrial's 65.7% → 63.0% decade movement is the only non-headline shift at decade level. The 2.7pp Industrial decline is smaller than the A/D +6.4pp rise, so the math does not let A/D alone account for the Industrial decline; A/D's share gain is also fed by smaller proportional pull-backs distributed across Education, Medical, and the residual.

This report describes the share movement and does not assert why defense rose or education fell. The why question is not derivable from AMPulse formation data alone. Report #6 in this series engages the question for the A/D rise directly.

Spotlights: eleven companies across the shift

Eleven companies, in two groups. Three era-illustrative companies whose formation pre-dates the 2010 cohort window but whose visibility defined the AM ecosystem the 2010s cohort formed within. Eight 2020s-archetype companies in our 2020-2026 cohort that exemplify the categories (Application VCP, Metal AM, Aerospace/Defense and adjacent end markets) toward which composition has moved. The cards are intended to give the abstract composition numbers a concrete reference image and are presented without commentary on outcomes.

Group A: Era-illustrative companies

These three companies were all founded before the cohort window opens in 2010 and are not strict cohort members. They appear here as the most-visible exemplars of the 2010s service-and-consumer AM ecosystem that the cohort numbers describe in aggregate. A fourth Group A seed (Voodoo Manufacturing, founded 2015) was originally planned as the only in-cohort era-illustrative card but is not present in the AMPulse database, likely pruned after Voodoo's 2018 shutdown. Voodoo's slot in the 2015-19 service-bureau-as-VC-bet archetype is therefore an unspotlighted reference.

Group B: 2020s archetype companies

Ordered for narrative flow: Pranos Fusion leads as the cross-reference to Report #1; KeratOPrinter closes as the most non-obvious card, requiring a deliberate "yes this is AM" reframing.

§ 4: 2024-26 micro-shifts

The 2024-26 cohort is small. N=163 over roughly 28 months of formation; the 2027-29 cohort is the one that will fully resolve whether the 2020s have settled or are still moving. With that caveat at the front, two within-2020s point-estimate moves show up in the 2024-26 data. Both are surfaced here as signals, not as findings.

The Hardware-VCP point estimate moved from 34.2% in the 2020-23 cohort to 39.9% in the 2024-26 cohort, a +5.7pp shift at the point estimate. Wilson 95% confidence intervals around these estimates are 30.9% to 37.6% (2020-23, N=766) and 32.7% to 47.5% (2024-26, N=163). The intervals overlap. We cannot reject the null hypothesis at 95% that the underlying Hardware share is unchanged between the two cohorts; the 2024-26 cohort is too small to distinguish "the share is unchanged and we drew a lucky sample" from "the share is genuinely rising further." The point estimate is consistent with the decade-aggregate Hardware reading of 35%, sitting above the long-run baseline but within the noise band a 163-company cohort would be expected to produce.

The Polymer-AM point estimate moved from 38.9% in the 2020-23 cohort to 46.0% in the 2024-26 cohort, a +7.1pp shift at the point estimate. Wilson 95% intervals are 35.5% to 42.4% and 38.5% to 53.7%. The intervals overlap again, by a wider margin than Hardware. Polymer's 2020-23 reading was the lowest in the four-cohort sequence (the steady decline visible in § 2); the 2024-26 reading is a partial reversal. As with Hardware, we cannot statistically distinguish "noise around the decade trend" from "an early sign the trend has bottomed and reversed." Either interpretation is consistent with the data we have.

Both signals deserve the same caveat: the 2024-26 cohort sits inside a still-open window. New companies founded in late 2025 and 2026 may not yet appear in the database; AMPulse's discovery pipeline takes time to surface recent formations. If the 2024-26 N rises meaningfully when re-extracted in 2027, both readings could shift by more than the current point estimates moved. The methodology box discloses this discovery-lag bias and notes that Report #3's funding-event audit will quantify it.

Two things the 2024-26 cohort does not show: any meaningful reversal of the Service-share decline (23.9% remains close to the 2020-23 reading of 25.2%, both well below the 2010-14 peak of 36.8%) and any meaningful reversal of the Aerospace/Defense rise (12.9% is essentially flat against the 2020-23 reading of 13.1%). The decade-level shifts the three earlier sections describe are not being unwound in the most recent cohort. What the 2024-26 cohort does show, within wide error bands, is a point-estimate re-concentration toward Hardware-and-Polymer. The 2027-29 cohort is what will resolve whether that is a real reversal of the 2020s composition or cohort-level sampling noise.

§ 5: what this signals

Three audience-tagged implications follow. They are forward-looking and necessarily softer than the descriptive §§ 1-4. Where §§ 1-4 say only what the data shows, this section says what to do with it. Readers who want to stop at the descriptive analysis should skip ahead to the closing asterisks.

Closing: the asterisks

The composition movement this report describes is real in the AMPulse database and is consistent across three classification axes (value chain, material, and target market) that describe different aspects of each company. The axes are not statistically independent (they share the same enrichment pipeline; see methodology), but they catch different facets of each company, which is why the same shift signal showing up on all three is more credible than any one axis alone. It is also subject to caveats worth stating directly.

• The shifts mostly resolved by 2020-23. Across all three axes, the 2020-23 → 2024-26 step is small or zero on every headline metric. Polymer's largest single-cohort step (−15.8pp) sits entirely inside the 2010s; Service's two largest steps span the 2015-19 boundary at roughly equal magnitude; A/D's largest single step (+4.9pp) crosses the 2015-19 → 2020-23 boundary. The 2020-23 cohort is where all three reached their decade-aggregate values. This report does not claim the shifts are currently accelerating. It claims they happened and the current cohort reflects the result.
• Formation share, not formation rate. The 2010-14 cohort has 1,318 companies in our database; the 2024-26 cohort has 163. This reflects some combination of genuine AM-formation-rate decline and AMPulse discovery lag for newer companies. The report uses share-only analysis throughout and does not claim formation-rate or industry-size trends. Report #3's funding-event audit will quantify discovery-lag bias when complete.
• Classifier evolution. AMPulse's material-category taxonomy was revised in April 2026 (Construction AM was promoted to its own top-level category; Novel Process was partially redistributed). Some 2010-14 companies are retroactively classified into categories that did not exist at their founding. The headline conclusions rest on the Polymer-vs-Metal core split, the seven-VCP structure, and the six-target-market structure (all stable across the update). Construction AM and Novel Process appear in the material chart but are not used as evidence for the thesis.
• Discovery bias. AMPulse's older-cohort enrichment may over-represent companies that were trade-press visible. If this bias systematically inflates the 2010s Service-and-Polymer-and-Education share, the magnitudes of the shifts reported here could be over-stated. The direction of the shifts is more confident than the exact percentage-point magnitudes.
• Subjective classification. Value-chain position, material category, and target market are AMPulse classifications. Reasonable analysts could re-classify and reach different numbers. The frozen JSON files committed alongside this report let any reader re-aggregate or re-classify against a different cut.

This report describes which categories the composition shifted toward and which it shifted away from. It does not assert why the shifts happened. The why question for each of the three movements is not derivable from AMPulse formation data alone. Single-axis deep-dives that engage the question explicitly are coming in Reports #5 and #6.

This is the second report in the AMPulse Data Reports series. Report #1, India's Sovereignty AM, covered the same database from a country-pair frame. Forthcoming reports in the series:

• R3, Series A Squeeze: capital bifurcation in AM funding rounds (in progress).
• R4, Active Investors 2026: who actually writes AM checks.
• R5, The Service Shop Bust: single-axis deep-dive on the Service-VCP decline.
• R6, Defense-Industrial AM: single-axis deep-dive on the Aerospace/Defense rise.

Send corrections to corrections@ampulse.online (or runsmlee@gmail.com until the alias exists).