The Factory Floor Nobody Imagined

Mass production without the robots.

When defense strategists talk about saturation - millions of drones, swarms of UGVs, mass over precision - the image is a lights-out factory. Robotic arms, CNC machines running overnight, production delegated entirely to software.

That is not what the factories look like.

A recent trip to Ukraine makes that clear. The shops producing land robots (“UGVs”) drones are small, often fewer than twenty people, assembling frames and soldering components by hand. Still, output per shop runs around 300 units a month without automation. This is not an isolate case and shows that scale can come from the design more than the assembly process.

Scale first, automate later

Anduril’s Arsenal-1 in Ohio is the most instructive example. When Fury (the unmanned combat aerial vehicle) production began in March 2026, the line had 22 workstations and no robots. The production floor, as one journalist who visited described it, resembled an Ikea more than a defense facility: bare bones stations, no large permanent tooling structures, nothing bolted to the floor that couldn’t be moved if the line needed reconfiguring.

Anduril’s Arsenal-1

At Arsenal-1, John Malone, Anduril’s head of production, came from Tesla, where he watched the Model 3 ramp nearly collapse under premature automation - cars ended up being assembled in a tent in a parking lot. His posture at Arsenal-1 is the inverse: run the line by hand first, and let it show you where automation is actually needed.

“When you go through early stage production, you’ll see what you should automate. It kind of presents itself to you,” John Malone, Anduril’s head of production.

Visiting Harmattan’s new factory last week, the same logic came up unprompted: production remains mostly manual today because designs are still changing too fast to justify locking in automated tooling. The moment you automate a specific assembly step, you’ve made it rigid. In a product still evolving, rigidity costs more than it saves.

Saronic embedded the same principle before building anything. Their head of manufacturing, recruited from SpaceX, was their third hire and joined before a single prototype existed. The reasoning: if production isn’t part of the design process from the start, whatever you build is only guaranteed not to scale. Their vessel platforms have 80% hardware consistency across the range and roughly seven core components per ship. The result is cheaper and more scalable products: their 24-foot Corsair costs under $1 million and can be produced at over 2,000 units per year.

Scale is a design decision

The reason human-heavy lines work at volume is that simplicity was engineered in upstream. Anduril chose abundant aluminum over rare titanium for the Fury airframe, standardized on commercial off-the-shelf components, and eliminated castings and forgings. Nearly 90% of Arsenal-1’s components are commercially sourced. The goal was an aircraft that automotive assembly workers, not aerospace specialists, could build consistently — a workforce decision made possible by a design decision.

Harmattan operates on the same principle. The ISR drone has four screws. The company minimized fasteners across the product line on the basis that every additional screw is assembly time and a potential failure point at scale. 90% of parts are shared across products, meaning a single production ramp covers the full portfolio. Like Anduril, Harmattan recruited deliberately from the automotive industry — people who think with a mass production mindset day one. The factory went from site identification to opening in 5 months, against an industry norm of 15, by doing almost everything in-house (as the design of the factory plan) and maintaining direct relationships with every supplier.

The Ukrainian shops I visited operate under a more extreme version of the same constraints. Output runs around 300 units a month per shop, assembled entirely by hand. But there is an additional variable that does not apply in the Western world: the factories move. Ukrainian drone producers operate across dispersed, constantly rotating sites because fixed facilities are targets for Shahed strikes. The production model has to be light enough to relocate on short notice - no heavy tooling, no fixed infrastructure, nothing that cannot be packed and moved. The reason it works at volume under those conditions is the same reason the model has to be mobile: the design is simple enough that humans can assemble it anywhere.

The automation question, in other words, is largely answered before anyone steps onto the factory floor. Assembly can stay human-heavy and still reach volume, provided the design decisions have been made correctly upstream.

Where software has the most to add

The assembly floor, in other words, is not where the industrial base bottleneck lives. The constraint is one level down, in the precision component shops that supply the assembly lines.

This $1.8 trillion industry is powered by small businesses whose owners are on average over 65, with 40% planning to retire within five years. That is where automation and software have the clearest value to add.

Hadrian, based in California, is building software-defined factories to address that gap. Its platform automates CNC tool path generation, quality inspection, scheduling, and adaptive programming, adjusting parameters in real time. A production cell goes from nothing to operational in under six months. New technicians can be trained in 30 days. As experienced machinists retire, the platform absorbs their process knowledge into the software layer. The company raised $260 million in 2025 at a $1.6 billion valuation with Founders Fund.

Isembard, started out from the UK, is working the same problem through a franchise model. It licenses MasonOS — an agentic AI platform integrating quoting, scheduling, supply chain management, manufacturing execution, and quality control — to existing machine shop operators and new entrants. The bet is that physical capacity already exists across thousands of small shops but what those shops lack is the software layer to run at defense-grade speed and consistency, which Isembard provides. Isembard raised $50 million recently with USV and is targeting 25 factories by end of year, spread across both the US and Europe. They’ve already signed an impressive list of clients, from Anduril, Tekever or Alloyed.

Isembard’s process

The two models differ — centralized automated factories versus distributed software franchises — but the diagnosis is the same. The factories that will determine whether saturation strategies are executable are not the ones producing the final product. They are the ones producing the parts.

---

Fundraisings:

• Hawkeye 360, a U.S. commercial satellite company that tracks radio-frequency signals from orbit, raised $416M in its IPO for a $2.4B valuation; shares rose 30% on the first day of trading.

• Anduril, a pioneer in autonomous military systems, raised $5B in a Series H at a $61B valuation, led by Thrive Capital and Andreessen Horowitz.

• Armada, building edge computing for military operations, raised $230M at $2B Valuation, led by Overmatch.

• Amca, a roll-up of critical component suppliers, raised a $300m Series B at a $1B valuation led by Caffeinated.

• HavocAI, the multi-domain robotics company known for its USVs, raised $100M in a Series A led by Cobalt Capital.

• SWEBAL, rebuilding TNT factories out of Sweden, raised €30M in a round backed by Sweden’s former defense minister Karl Engelbrektson, ex-EQT CEO Thomas von Koch, and several Swedish family offices.

• Darkhive, a tactical autonomy and edge software company built around small UAS, raised $30M in a Series B led by RTX Ventures.

• Heaviside, building autonomous precision munitions, raised $28M in a Series A led by Interlagos Capital.

• Anello Photonics, a silicon photonic optical gyroscope company, raised $25M in a Series B-2 led by MESH. .