Industries / Industrial equipment and services / 3D printing
Metals
Metal printing builds parts layer by layer from metal powder or wire. The metal is fused with a laser, an electron beam or an electric arc, or held together with a liquid glue and then baked solid in a furnace.
7
Milestones achieved, from 1995 to 2024.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1995
Lasers fuse metal powder
EOS, a German laser-machine maker, introduced the EOSINT M 250, which its own timeline calls the first series-production machine for direct metal laser sintering (DMLS: a laser traces each layer in a thin bed of metal powder). EOS says it marked the start of "rapid tooling" (making metal molds and tools quickly), and it also lists an earlier machine, the EOSINT M 160, in 1994. The metal process grew out of work at Electrolux's prototyping unit in Finland, where engineer Olli Nyrhilä began melting metal powder with lasers. In 1996 three researchers at the Fraunhofer Institute for Laser Technology ILT in Aachen filed a patent on melting metal powder layer by layer with a laser, the process now called laser powder bed fusion (LPBF), which by ILT's 2021 count made up over 80 percent of the metal printing market.
LessonA patent fight can keep a working machine out of a market for years, and by TCT Magazine's account EOS sold no metal machines in the US until 2004 because of drawn-out litigation.
eos.infotctmagazine.com3dprintingjournal.comilt.fraunhofer.de
Credit
EOS (founded by Hans J. Langer); Electrolux Rapid Prototyping (Olli Nyrhilä); Fraunhofer ILT (Wilhelm Meiners, Kurt Wissenbach, Andres Gasser)
1997
Powder sprayed into a laser
Sandia National Laboratories in New Mexico developed Laser Engineered Net Shaping (LENS), a form of directed energy deposition: nozzles aim streams of metal powder at a spot heated by a laser, and the part is built up layer by layer from a computer model. The project started in 1995. In December 1997 Sandia announced a two-year, $3 million agreement with ten companies, among them Lockheed Martin, 3M and Optomec, to turn it into a commercial machine, with an estimated price of $350,000 to $500,000. Optomec of Albuquerque licensed LENS in 1997 and has been its only maker since, according to a University of California, Irvine lab page.
Credit
Sandia National Laboratories (project manager Clint Atwood), with early work alongside Pratt & Whitney; Optomec
1998
Glue first, furnace later
At MIT, Emanuel Sachs and Michael Cima invented binder jet printing, in which an inkjet printhead drops liquid binder into a bed of powder, and Sachs coined the name "3-D printing" for it. Extrude Hone, a US firm, set up a ProMetal division in 1995 and took an exclusive field-of-use license to MIT's patents in 1996. ExOne's later brochure says the ProMetal RTS-300 of 1998 was the first metal printer to use binder jetting, but ExOne's 2013 share prospectus says Extrude Hone made its first 3D printing machine in 2003. Printed metal parts need curing and sintering (heating until the grains bond), and ExOne says it began printing full-density single-alloy parts without infiltration (filling the pores with a second metal) in 2013.
Credit
MIT (Emanuel Sachs, Michael Cima); Extrude Hone, later ExOne (founder Larry Rhoades)
2003
Electron beams in a vacuum
Arcam, founded in Sweden in 1997 from work begun at Chalmers University of Technology in Gothenburg, sold its first electron beam melting (EBM) machine in 2003, to North Carolina State University. EBM melts the powder with a beam of electrons steered by magnetic coils instead of mirrors, inside a vacuum, and heats each powder layer to about 1,000 °C first, which by the account of Colibrium Additive (GE's printing business, which bought Arcam in 2016) leaves parts with very little built-in stress. Arcam's team had first tried to print with an electric arc, as in welding, but could not hit the same spot with the same energy twice, so it switched to an electron beam.
LessonLaser printing of titanium aluminide turbine blades failed from thermal stress, and by Colibrium's account the hot powder bed of EBM is what made them printable.
Credit
Arcam AB (from master's work by Morgan Larsson at Chalmers)
2021
Robots print a steel bridge
MX3D, a Dutch company, used four industrial robots to print a stainless steel footbridge for an Amsterdam canal from over 6,000 kg of steel, by 3D Printing Industry's report. The method is wire arc additive manufacturing (WAAM: a form of directed energy deposition in which a welding arc melts metal wire bead by bead), and MX3D's software turns off-the-shelf robots and welding power sources into metal printers. Printed in 2017 and 2018, designed by Joris Laarman Lab and engineered by Arup, the bridge was opened by Queen Máxima in July 2021 with sensors feeding a digital model of its behavior. The City of Amsterdam gave it a two-year permit, and it was removed in October 2023 to be set up at a new site.
LessonPrinted infrastructure entered public use as a time-limited, sensor-watched trial rather than as a permanent structure.
Credit
MX3D (co-founder Tim Geurtjens), Joris Laarman Lab, Arup, City of Amsterdam
2023
A printed rocket, then retreat
On 22 March 2023 Relativity Space launched Terran 1 from Cape Canaveral; about 85 percent of its 9,280 kg structural mass, including its Aeon engines, was printed, by the company's figures as reported by Spaceflight Now. The printed structure came through the point of maximum aerodynamic pressure, but the second-stage engine did not reach full thrust and the rocket did not reach orbit. Three weeks later Relativity retired Terran 1 and said its larger Terran R would use a "hybrid manufacturing approach," including conventionally made aluminum tank sections, with printing used "strategically," CNBC reported.
LessonPrinting a whole large structure was not the fastest route to a working vehicle, so Relativity kept printing where it cuts complexity and went back to conventional methods for simple tank sections.
Credit
Relativity Space (CEO Tim Ellis)
2024
China leads machine shipments
In the first quarter of 2024, four of the five vendors that shipped the most industrial metal powder bed fusion machines worldwide were based in China (Eplus3D, BLT, ZRapid Tech and Farsoon), with Eplus3D first, according to the market analyst CONTEXT. Shipments from Chinese vendors rose 45 percent year on year while Western vendors' fell 4 percent, though Nikon SLM Solutions and EOS still led on machine revenue. Chinese makers are known for very large multi-laser machines: Eplus3D said in January 2025 it had sold over 100 "super-meter" machines, nearly 40 of them over a meter on all three axes, including the 36-laser EP-M2050 with a 2,050 × 2,050 × 1,100 mm build volume.
Credit
Eplus3D, Bright Laser Technologies (BLT), ZRapid Tech, Farsoon Technologies; measured by CONTEXT
Watch for
A 3-meter, 100-laser machine
Eplus3D launched the EP-M3050, a laser powder bed fusion machine with a build area 3 m on each side that starts with 100 lasers and can be scaled to 256, for titanium, aluminum and steel parts; Fabbaloo reports the system can scale up to 5 × 10 × 5 m.
Achieved whena customer outside Eplus3D reports an EP-M3050 installed and printing, or an outside party reports a completed full-size build with its print time.
Claimed by
Eplus3D (China), April 2026
Machine-learned printable aluminum
MIT researchers used machine learning, evaluating 40 compositions instead of simulating over a million, to design an aluminum alloy for laser powder bed fusion. By the team's own measurements, published in the journal Advanced Materials, printed samples were five times stronger than a cast counterpart and 50 percent stronger than alloys designed by conventional simulation alone, and the alloy's structure stayed stable up to 400 °C.
Achieved whena lab outside the author team prints the alloy and reproduces its strength, or a powder supplier sells it and a customer reports printed parts.
Claimed by
MIT (Mohadeseh Taheri-Mousavi, now at Carnegie Mellon, and John Hart) with Paderborn University and Carnegie Mellon co-authors, October 2025
Open
01
Standing scorer for melt-pool models
A public scoring site and script set, open all year, that takes a simulation's predicted melt-pool width and depth (the size of the small pool of liquid metal under the laser) for the NIST AM Bench single-track and pad test cases and reports the error against NIST's measurements, with simple baseline entries for others to beat.
The briefCredit · unclaimed02
Claimed versus measured build rates
An open table that lists, for each metal printing machine, the maker's stated build rate (cubic centimeters of metal per hour) next to rates worked out from independently documented builds (papers, customer case studies or published build logs that give part volume or mass and print time), with the ratio between the two and the method used.
The briefCredit · unclaimed
Open 01 · the brief
Standing scorer for melt-pool models
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public scoring site and script set, open all year, that takes a simulation's predicted melt-pool width and depth (the size of the small pool of liquid metal under the laser) for the NIST AM Bench single-track and pad test cases and reports the error against NIST's measurements, with simple baseline entries for others to beat.
Why it is still open
NIST runs AM Bench as rounds of blind modeling challenges tied to conferences (2018, 2022, 2025, with 2028 planned); the 2025 round drew 86 submissions from 24 groups, and awards were given per challenge, including melt-pool geometry challenges. Searches of NIST's pages, GitHub, Hugging Face, Kaggle and Zenodo in November 2026 found no public place to score a new model between rounds. NIST's GitHub holds data releases and 2025 submission templates, not a scorer. MeltpoolNet from Carnegie Mellon collects melt-pool sizes from published papers; a September 2026 project, meltpool-generalization, publishes fixed test splits of that literature data, not AM Bench measurements; and ORNL's open AdditiveFOAM solver ships an AM Bench 2018 tutorial case. None scores outside models against AM Bench measurements on a public leaderboard. Machine-learning melt-pool models are now common, which is why a shared scorer is worth having.
Built with what exists
NIST AM Bench public data, ORNL AdditiveFOAM (open source, GPL-3, built on OpenFOAM), the MeltpoolNet CSV datasets, Python, and a GitHub repository with automated scoring (for example GitHub Actions) or a Hugging Face Space for the leaderboard.
First step
Run AdditiveFOAM's AMB2018-02-B tutorial this month and compare its melt-pool widths with NIST's published measurements for that case.
Done looks like
Scoring scripts and data links published under an open license; at least three baseline models (for example a simple analytic heat-source model, an AdditiveFOAM run and a model trained on MeltpoolNet) scored on at least two AM Bench melt-pool cases, with error in micrometers; at least one group from outside the builders submits an entry.
Open 02 · the brief
Claimed versus measured build rates
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open table that lists, for each metal printing machine, the maker's stated build rate (cubic centimeters of metal per hour) next to rates worked out from independently documented builds (papers, customer case studies or published build logs that give part volume or mass and print time), with the ratio between the two and the method used.
Why it is still open
Makers publish their own speed figures, such as Eplus3D's 1,080 cm³/h for its 36-laser EP-M2050 as reported by Fabbaloo, and the largest machines now carry 100 lasers. The free Senvol Database lists industrial machines and materials but says Senvol is not liable for the accuracy of its information and tells users to check with the makers, and CONTEXT counts shipments, not speeds. Searches on wire-arc deposition rates found vendor and research figures measured under different definitions that cannot be compared directly. The closest independent work found, a 2022 Barnes Global Advisors test of one electron-beam machine, looks at its economics and gives no build rates. Searches in November 2026 found vendor claims and single-machine measurements but no public table that sets claimed rates against measured ones.
Built with what exists
Makers' datasheets, the Senvol Database, published papers and case studies, PySLM (an open Python library for laser powder bed build files that includes build-time estimation from scan strategy and geometry) to estimate times for a reference part, and a CSV file on GitHub.
First step
Collect the stated build rates for ten large laser powder bed machines from datasheets and the Senvol Database, and find one published build with part size and print time for each.
Done looks like
Twenty machines across at least three processes (laser powder bed, electron beam, wire or powder deposition), each with the maker's figure, at least one independent build with part size and print time, the computed ratio, and a written method; published as an open CSV.