By Staff
The old rules of making things are breaking down. A network of factories across the United States is building a manufacturing base that doesn't depend on Chinese casting supply chains, doesn't need tooling that takes months to produce, and doesn't care whether a part is too complex to exist. The Pentagon is writing the checks, AI is designing the parts, and the energy problem is solvable. Here's what's actually happening.
In a 430,000 square foot facility in Long Beach, California, there are no furnaces, no molten metal, no sand molds. Instead, rows of machines fire lasers into beds of metal powder, fusing titanium and nickel alloys layer by layer into shapes that no casting process could produce. Divergent Technologies calls its in house built machine the Monolith One, and it was designed from the ground up for one purpose: producing tens of thousands of munition air frames and hundreds of thousands of critical components per year.
Six of these printers are already running at the company's Torrance headquarters. Sixty four more are slated for Long Beach over the next two years. When Secretary of War Pete Hegseth toured the facility in January as part of his Arsenal of Freedom tour, the message was clear. The Pentagon is betting that the future of defense manufacturing looks less like a traditional foundry and more like a data center full of printers.
Divergent is not alone. Across the country, a constellation of facilities is standing up production scale metal additive manufacturing and nearly all of it is underwritten by defense dollars.
In Knoxville, Tennessee, and Centennial, Colorado, Beehive Industries is running one of the largest metal AM operations in the country, printing engine components for its Frenzy 8 engine line and Rampart turbofan platform. The company just dropped fifty million dollars on thirty new EOS M4 ONYX systems, doubling its capacity. These are six laser machines with advanced process monitoring, designed for serial production, not prototyping.
In Cleveland, Ohio, Lincoln Electric the welding company most people associate with stick electrodes and hard hats is running twenty six large format robotic Wire Arc Additive Manufacturing systems around the clock. Their WAAM cells handle components up to twenty thousand pounds, printing parts that would otherwise require castings or forgings with lead times measured in months or years. Lincoln acquired Baker Industries in Michigan and built a fully vertical pipeline: print, machine, fabricate, inspect. They are quietly eating the casting industry's lunch on large industrial components.
In Houston, Texas, DEEP Manufacturing has a fifty thousand square foot facility running four WAAM systems with more coming, focused on pressure rated vessels and large components for the energy, defense, and maritime sectors. They are chasing DNV certification for subsea pressure vessels a world first for the process. Houston puts them in the backyard of the oil and gas industry, where replacement parts for offshore platforms have brutal lead times through traditional casting supply chains.
In Burgettstown, Pennsylvania, 6K Additive is not printing parts but producing the metal powder that feeds the printers and doing it with a proprietary UniMelt process that bypasses traditional gas atomization entirely. Currently at two hundred metric tons of annual capacity, they are expanding to a thousand metric tons with a twenty three million dollar Defense Production Act Title III grant. They are making nickel, titanium, stainless steel, and refractory metal powders including tungsten, rhenium, and C-103 alloy for hypersonics. The Pentagon is funding half the expansion because domestic powder supply is a national security vulnerability. Too much of it currently comes from countries the United States may not want to depend on for missile alloys.
In South Boston, Virginia, IperionX just commissioned a three hundred ton six axis powder metallurgy press at its Titanium Manufacturing Campus, tripling its compaction capacity. Their entire play is a fully domestic titanium supply chain from US sourced feed stocks through a patented process for powder production, to near net shape pressing and sintering. That press can crank roughly eleven million titanium parts per year. The Department of Defense has a contract worth up to forty seven million dollars to build out the full domestic titanium chain from mineral extraction through finished components.
The timing is not accidental. The US shed most of its foundry base over the past forty years. The casting supply chain for critical alloys is concentrated overseas. When you need replacement parts for weapons systems in active combat and the traditional pipeline runs through countries with complicated geopolitical posture, additive manufacturing stops being a cool technology demonstration and starts being a strategic necessity.
The Iran war that began in March 2026 only intensified the urgency. But the structural drivers predate any single conflict. The Department of Defense has been reading the tea leaves on supply chain fragility for years. The Defense Production Act grants, the EXIM loans, the Defense Logistics Agency contracts to convert military grade scrap into high performance powder these are not market-driven investments. They are industrial policy executed through the Pentagon's budget.
The economics are shifting too. For decades, the rule was simple, if you needed more than a thousand units, casting was cheaper. The break even between additive manufacturing and casting has historically sat somewhere between sixty and a thousand units depending on part complexity. Below that, AM wins because it eliminates tooling costs that can run from five thousand to over a hundred thousand dollars for a single mold. Above that, the amortized tooling makes casting the dominant economic choice.
But that analysis assumed a human driven design process. Inject artificial intelligence into the front end and the equation changes not dramatically on per part cost, but radically on speed.
Generative design is the part everyone talks about. Feed an AI load cases, boundary conditions, material properties, and manufacturing constraints, and it produces a topology optimized geometry no human would conceive organic lattices, variable density infill, load paths that resemble bone structures. The AI designs for the printer, and the printer produces what the AI imagines. That eliminates the traditional back and forth where design engineers create something manufacturing engineers have to beat into feasibility.
But the real time savings are less glamorous and more consequential. Once you have a design, a skilled AM engineer might spend a day or two on build preparation orienting the part in the build volume, generating support structures, setting laser parameters, simulating thermal distortion and compensating for it. AI driven tools can do this in minutes. More importantly, they can iterate. The AI tries thousands of orientation and support configurations, runs thermal simulations on each, and converges on the build strategy that minimizes distortion, support volume, and print time simultaneously.
The fully integrated pipeline looks like this: requirements go in, AI generative design produces a geometry, AI build prep optimizes the manufacturing strategy, the printer runs, in process monitoring with cameras and melt pool sensors feeds data to AI defect detection, and that data loops back to improve the next build. The iteration cycle collapses from weeks to hours.
Modern powder bed fusion machines watch every layer with cameras and photodiodes. AI vision systems flag anomalies porosity, spatter, lack of fusion in real time. That data does not just reject bad parts. It feeds forward into the design and build prep for the next part. Instead of design, print, inspect, find problem, redesign, print, inspect, certify, you get design, print with real time AI monitoring and correction, certified part.
This also changes who can run these machines. Right now, the bottleneck is finding people who know how. A skilled AM engineer commands a salary north of a hundred and twenty thousand dollars and is hard to find. AI build prep and automated workflows reduce the skill floor. A technician with six months of training can oversee a cell that previously required an engineer with five years of experience. That is great for scaling. It is less great for the engineers who spent years building that expertise.
There is a deeper issue the industry does not discuss openly, who controls the AI that controls the printers? If your build prep AI was trained on data from a specific manufacturer's machines running that manufacturer's parameter sets with that manufacturer's approved powders, you are locked into that ecosystem. The democratization of AM through AI could become a new form of vendor capture easier to use, harder to escape.
For aerospace and defense, certification is the long pole. AI generated designs and AI-optimized build strategies create a validation headache. How do you certify a part when the design and manufacturing process are both non deterministic from a human perspective? The FAA and DoD qualification frameworks were built around the assumption that a human engineer can explain why they made every decision. When an AI converges on a geometry through thousands of iterations of a generative algorithm, the why becomes inscrutable. You can validate the output test it to destruction, inspect it with CT but you cannot trace the design logic. The agencies are working on this, but it is a genuinely hard problem that is not solved by throwing more compute at it.
For all its advantages, metal additive manufacturing has an energy problem. Producing a part via laser powder bed fusion consumes five to ten times more energy per kilogram than casting. The lasers or electron beams draw enormous power, the gas atomized powders carry high embodied energy from their own production, and post-processing steps like hot isostatic pressing add another massive thermal cycle.
But the problem is solvable, and multiple attack vectors are in play.
Solid state processes like Additive Friction Stir Deposition skip melting entirely. They use severe plastic deformation to bond material without a melt pool, cutting energy consumption by sixty to eighty percent compared to powder bed fusion. MELD Manufacturing in Virginia has been pushing this for large structural parts. It cannot match the fine feature resolution of laser systems, but for the kind of big chunky components that currently get cast, it is a direct energy competitor.
Binder jetting takes a different approach. Instead of melting powder with a laser, an inkjet head deposits a liquid binder onto a powder bed at room temperature. The green part then goes into a furnace for sintering. The furnace step is energy intensive, but you sinter thousands of parts simultaneously in one batch, amortizing that energy across the whole load. Desktop Metal and HP have been driving this approach. The trade off is material properties sintered parts are not quite as dense as fully melted ones but for non structural applications, the energy math is compelling.
Multi laser systems attack the problem from a different angle. A single laser running for eighty hours on a large build loses enormous energy to heating the chamber, the recoater, and the build plate. Systems with four, eight, or sixteen lasers finish the same build faster, which means less total chamber heating time and less energy lost to surroundings. The EOS M4 ONYX that Beehive is installing uses six lasers. Chinese manufacturers are already fielding sixteen laser behemoths. Faster builds equal less energy per part.
Eliminating hot isostatic pressing could be the single biggest lever. HIP is standard for aerospace parts load them into a pressure vessel, heat to near melting under enormous argon pressure for hours. It closes internal porosity but at staggering energy cost. If you can print dense enough through better process control, AI driven parameter optimization, and real time melt pool monitoring, you eliminate HIP. The FAA and DoD qualification frameworks are slowly warming to as printed or heat treat only pathways for non-rotating structural parts. Every HIP cycle you skip saves more energy than optimizing the printer itself.
Further out, alternative powder production routes change the upstream equation. A huge slice of AM's lifecycle energy is in the powder, not the printer. Gas atomization melts the alloy, holds it at temperature, and blasts it with high pressure argon. Electrolysis based processes that go directly from oxide to powder without melting, plasma spheroidization of lower cost irregular powders, and recycling of machine shop swarf into usable powder all cut the embedded energy before the material ever reaches a printer.
Nuclear colocation is not as far fetched as it sounds. A factory running seventy high power printers around the clock is exactly the kind of steady, high density electrical load that makes small modular reactor economics work. Dow is already planning SMRs for chemical processing. Defense manufacturing campuses with dedicated micro reactors are being studied. The printer does not care where the electrons come from. Cheap, clean base load power changes the economics of an electricity dominant manufacturing process entirely.
The holy grail is ambient temperature deposition. Electro chemical additive manufacturing deposits metal ions from solution onto a substrate at room temperature. No heat, no powder, no laser. The energy input is the electro chemical potential orders of magnitude lower than melting. The catch is that deposition rates are currently glacial, the alloy range is limited, and the material properties are not there yet for structural applications. But the physics is sound. If someone cracks the rate problem, the entire AM versus casting energy debate becomes irrelevant.
The United States is not alone in this race. China's Jingye Additive the same Jingye Group that bought British Steel in 2020 operates forty-six metal AM machines including six so called super-meter class systems with sixteen lasers. They have delivered over a hundred thousand parts and have sunk more than seven hundred and fifty million dollars into additive manufacturing. Anyone who thinks the US holds an unassailable lead is not paying attention.
But the American advantage may lie less in hardware than in the software stack. The AI driven design to manufacturing pipeline generative design, automated build preparation, in-process monitoring, closed loop quality control is where US companies and defense contractors are concentrating investment. China can build printers and produce powder. The question is whether the integrated digital thread from requirements to certified part becomes the real competitive moat.
Meanwhile, the sustainability comparison between AM and casting is more nuanced than either side's advocates admit. AM achieves ninety to ninety-five percent material utilization compared to casting's fifty to seventy percent. It generates five to ten percent waste versus thirty to fifty percent for casting. Unused powder can be sieved and reused. But the energy intensity is five to ten times higher per kilogram processed.
The honest assessment is that AM is greener when it enables a design that could not exist otherwise a topology optimized aircraft bracket that weighs forty percent less and saves thousands of tons of jet fuel over the aircraft's life, or a consolidated engine component that replaces an assembly of twenty individually cast and machined parts. For simple solid geometries at high volume, casting remains the lower carbon option. The environmental winner depends entirely on what you are making and how many you need.
The metal AM plants operating today are real, they are running, and they are producing. But operational does not mean profitable without government support. The economic case for AM at production scale still depends heavily on parts that are impossible or absurdly expensive to cast, defense programs where cost is secondary to supply chain sovereignty, and replacement parts for legacy systems where tooling no longer exists.
The dream of AM replacing casting for commodity production remains a fantasy. But the factory of 2030 will not look like the factory of 1980. It will have both casting cells and banks of printers. The competitive advantage will go to whoever figures out how to make them work together casting the simple bulk, printing the complex features, using AI to optimize across both processes simultaneously.
The foundry is not dead. But it is no longer the only way to make metal parts that matter. And for a country that let most of its foundries disappear, that may be enough.
Sources:
ISO/ASTM 52900 (AM terminology / process categories): amazemet.com
ASTM/ISO-based “additive manufacturing file format” (AMF, ISO/ASTM 52915): twi-global.com
Examples of AM material/process qualification standards for metal AM (e.g., Ti-6Al-4V with powder bed fusion; nickel alloys with powder bed fusion): twi-global.com
Overview of major metal AM process types (PBF / DED / Binder Jetting): amazemet.com1
Divergent Technologies — June 17, 2026 press release announcing Monolith One and the Long Beach factory (PRNewswire)
Beehive Industries — June 17, 2026 coverage of their $50M EOS M4 ONYX investment (TCT Magazine)
DEEP Manufacturing — March 27, 2026 coverage of the Houston WAAM facility opening (3D Printing Industry)
6K Additive — April 30, 2026 coverage of the Burgettstown expansion and DPA Title III grant (3D Printing Industry)
IperionX — May 21, 2026 press release on SACMI press commissioning in Virginia
Lincoln Electric — March 17, 2026 coverage of their 26 WAAM systems and casting replacement strategy (3DPrint.com)
Jingye Additive — November 2023 coverage of their super-meter-class printer fleet (3DPrint.com)
