← Darius Foodeei

The last call to prosperity

· Darius Foodeei

The bet

Every civilization is, at it's core, a machine for turning matter into a way of life. The rights we cherish, the hospitals we trust, the pensions we were promised, the freedom to argue and to leave: all of it rests on a pile of steel, copper, silicon and energy that someone had to dig up, refine, shape and put together.

For two generations the West decided that someone could be someone else. We kept the design, the brand and the margin, and we sent the making away. It looked like wisdom. It was a loan, and an unexpectedly heavy bill has arrived.

This is an optimistic essay. I believe the next fifty years can be the greatest era of building the West has ever seen. But optimism is not comfort. The bet is simple: either we automate every step from the mine to the finished product, with machines each built for one job and an intelligence they all share, or we become a beautiful museum of our own past. Admired, visited, and irrelevant.

The good news is that we already know how to win. Most of the tools were invented here. We only have to remember that making things is not beneath us. It is the foundation of everything above us.

The ledger

We are not losing everywhere. We are losing at the bottom of the stack, which is the worst part of the stack to lose, because everything else stands on it.

Layer The West China
Rare-earth refining A handful of plants, mostly still being built About 91% of global refining and processing (S&P Global, citing IEA)
Factory robots installed, 2025 Germany, Europe's largest market: under 25,000 354,000, or 59% of the world (IFR World Robotics 2026)
Ships delivered, 2024 US: 28 ships, 0.04% of global tonnage 1,286 ships, about 55% (Clarksons via Yicai)

In April 2025, Beijing placed export controls on seven heavy rare earths and their magnets. Within weeks, car lines in Europe and America went quiet. In July 2026 it barred 14 European companies from buying them at all. No blockade, no missile. A permit office was enough.

That is the quiet lesson of this decade. Dependence is invisible until the day someone decides to use it. And the advantages we still hold, the frontier models, the great universities, the deepest capital markets, the machine-tool makers of Swabia and the watchmakers of the Jura, are not a fortress. They are a head start and quickly expiring

Yet none of this is destiny. China did not get here by magic. It got here through decades of patient, deliberate, unglamorous engineering. That used to be our specialty.

Built for one job

I have spent long nights tuning policies on humanoid robots. I love them. They are among the most beautiful engineering problems of our time, and the research they drive will push physical intelligence forward for decades.

But we should be honest about what they are not. They are not a reindustrialization strategy.

The human form is a compromise. It is the shape that can do a little of everything in a world built for humans. That makes it seductive, and the wrong unit for for an industrial strategy. A factory does not need one body that can do everything. Human bodies in factories are flexible, but inefficient and expensive for singular repetitive operations. Humanoids duplicate those costly trade-offs. A high-throughput factory needs many specialized bodies that each execute one task with extreme efficiency, speed, and precision.

Computing already learned this lesson. We never built one universal chip. We built GPUs for training, accelerators for inference and microcontrollers for washing machines, each shaped for its job. What we share across all of them is the software. The same model can run in a datacenter, a phone and a car.

Robotics should make the same split.

My conviction is simple: robots should be built to their purpose, and intelligence should be shared. The body of a machine should be shaped by its task: a gripper cut for one connector, a press for one bearing, a six-arm cell for one drone. Its mind should come from a common pool of physical intelligence that learns from every cell in every factory and makes all of them better at once. A gripper should not have to be a hand to benefit from everything hands have learned. Specialized bodies, shared brains.

The economics point the same way. A Unitree G1 carries 23 to 29 actuators before you count its hands; a six-axis arm has six. Actuators are the bulk of a humanoid's bill of materials, so every motor a task does not need is money, mass, energy and one more part that wears out. A stamping press runs for years at thousands of strokes an hour on a few actuators. And a machine that does one thing is far easier to validate. Its failure modes can be listed, and a list of failure modes is what a plant manager actually buys.

None of this makes general-purpose robots useless. It gives them a different job. They take on tasks while those tasks are new and rare. Once a task becomes frequent and stable, a purpose-built machine takes over, and the generalist moves on, often helping build its replacement. Generalists are the scouts. Specialists are the economy, and that is where almost all the value lives.

What changes everything is the cost of making a specialist. A specialized machine only pays for itself if it stays busy, which is why today it is reserved for products made by the million. When AI designs the machine and robots build it, that threshold falls from millions of units to thousands, then hundreds. The result will not be a world of humanoids. It will be an explosion of machines, each built for exactly one job (a point made well in the essay The real AGI-pill is specialized robots).

The builders are already moving this way. Travis Kalanick's Atoms builds what it calls "gainfully employed robots": specialized machines with real jobs, from autonomous haul trucks in mines to kitchens that cook at industrial scale. It argues a humanoid is the worst possible way to make a thousand pancakes an hour. Atomic Machines goes further, compiling working micro-machines directly from code. Hadrian is rebuilding American precision manufacturing with AI-run factories that read designs and automate machining and inspection. Three very different companies, one conclusion: the machine should fit the job.

A country that bets its industrial future on general-purpose robots alone is hoping, not building. China is doing both. It can afford to experiment with humanoids because it already owns the factories.

We already know how to do this

Here is the optimistic part: The hardest problems in reindustrialization are not mysteries but rather optimization problems, and the West invented optimization.

Linear programming, operations research, finite elements, numerical control, control theory, the CNC machine, the transistor. These were built in places like Princeton, Cambridge, Zurich and Paris. Reindustrialization does not need a miracle. It needs us to point our best tools back at the physical world, at every step of the chain:

  1. Extraction. Autonomous drills and haul trucks already run entire mines. What is missing is the will to open deposits at home, and permits measured in months rather than decades.
  2. Refining. The least glamorous and most strategic layer. Rare-earth separation runs through hundreds of chemical stages in series, each one a control problem. A plant that models its whole cascade end to end can run tighter, cleaner and smaller than one tuned by hand.
  3. Forming and machining. Already largely programmable. The next gain is closing the loop: machines that sense their own wear, drift and defects, and correct themselves mid-run.
  4. Assembly. Where most of the remaining human hours go, because it is contact-rich, high-variance and different for every product. Hand-programmed motion breaks down here. This is where learned physical intelligence earns its keep: inserting, fastening, routing a cable, feeling the moment a part is seated. Physical AI alone cannot do this at scale but is an essential part of the optimization problem.
  5. Design. Where the loop closes. Every problem a machine solves should flow back into how the next product, and the next machine, is drawn.

Deterministic lines, however clever, are re-engineered by hand for every new product. Pure robot learning, however impressive, has no idea where an arm should stand, how a line should flow or what the product should look like. Neither alone automates industry at scale.

The breakthrough I believe in is the combination. Treat the whole production system as one optimization problem: the product's design, the shape of each robot, the layout of each cell, the flow of the line. Optimization decides what to build and where. A shared intelligence handles what no plan can anticipate: the connector that has to click, the thread that catches, the part that sits half a millimeter off and must be nudged home.

That intelligence is shared across every machine. Each shift any robot works teaches all the others about contact, materials and skills, while each body keeps its own kinematics and calibration. And what the machines learn on the floor flows back into the optimizer, reshaping how the next product, the next cell and the next robot are designed. The CAD file becomes the start of a loop, not the end of a handoff.

This radical simplicity is what makes extreme cheapness possible and extreme cheapness is the core engine of reindustrialization. Optimization strips each body down to the simplest mechanism its task requires: three axes instead of seven, a jaw cut for one part instead of a hand, off-the-shelf motors, a frame you can machine in an afternoon. Shared intelligence absorbs the variance that simple hardware cannot. Machines this simple are cheap to build, cheap to repair, and simple enough to be built by other machines.

That is where AI and frontier physical intelligence truly matter. Not as a replacement worker, but as both halves of one system: the optimizer that designs the factory, and the shared mind that lets a thousand simple machines inside it behave like experts. The target is a capable robot that costs less than a year of the labor it replaces. When we reach it, labor cost differentials between countries no longer dictate where goods are manufactured.

This is what hyperscale manufacturing means: producing purpose-built machines as easily as we produce software. When bodies are cheap and minds are shared, we stop bending every task to fit one form and start shaping the form around each task.

This is what I am setting out to build: industrial automation at scale, where every robot is designed for its task, costs little, and gets smarter every time any of them works a shift. It is hard. It is also exactly the kind of hard we are good at.

Sovereignty, 2026 edition

2026 is the year the West rediscovered that intelligence has a physical address. Models run on chips, chips run on power, and both sit somewhere on a map.

The signs are everywhere this year:

This week Arthur Mensch put the stakes plainly in Build or Submit. European companies already pay more than €260 billion a year to American cloud and software providers, and whoever owns a model sets its terms (Cybernews). In June, US export controls briefly suspended access to the most capable American models; access was restored on 1 July. Three weeks was enough to make the point. Even between friends, dependence is a variable someone else controls.

But language models are only the first act. The next frontier is physical intelligence: models that understand force, contact, materials, tolerances and throughput. And here is the uncomfortable truth. That intelligence is trained on data that only factories produce. Every robot that runs a shift generates the experience the next model learns from.

That makes the loop brutal. Factories produce the data for physical AI, and physical AI produces the factories. China installed 354,000 industrial robots in 2025. Each one is also a sensor. A West that builds sovereign chatbots but not sovereign production lines will own the conversation and rent the world.

Rivals in the same cause

I want the nations of the West to compete ferociously. Competition is how we got the jet engine, the moon landing and the smartphone. Let Bavaria and Texas race to build the most automated battery line. Let France and Arizona fight over the next fab. Airbus and Boeing made each other better for fifty years.

But let us be clear about what we are competing for. Not to defeat one another. To prove whose path to a shared prosperity is fastest. The rivalry should look like a race between teammates, not a war between neighbors.

The most advanced machine humanity builds is the proof. An EUV lithography system is assembled in the Netherlands by ASML, sees through German optics from Zeiss, and fires a light source engineered in San Diego. No single Western country could build it. Together, nobody else can.

That is the model. America brings scale, capital, frontier labs and the courage to bet big. Europe brings precision engineering, deep industrial know-how, a dense web of family-owned manufacturers and the patience to make things last. Neither can reindustrialize alone. Both can, together, in a fraction of the time.

So here is what alliance should mean in practice: shared standards for robots, data and interfaces; allied supply chains for minerals and magnets; open access to each other's compute; and joint stockpiles of the things a permit office can take away. Tariff fights between allies are family arguments. We cannot let them become a divorce, because the only winner of a divided West is whoever is building while we argue.

The next factory floor has no floor

Within fifty years, the most important industrial region of the solar system may not be on Earth. Orbit will need power plants, foundries and shipyards of its own. The Moon has oxygen, metals and water locked in its soil. Asteroids hold more platinum-group metals than humanity has ever mined.

Space is the ultimate test of everything in this essay. There is no cheap labor in orbit. There is no labor at all. A signal takes over a second to reach the Moon and up to twenty minutes to reach Mars. Every step, from extraction to refining to fabrication to assembly, must run on its own, fix itself, and adapt to parts that were never quite what the drawing said.

Space is also where specialization stops being an optimization and becomes a requirement. Every kilogram sent to orbit costs a fortune, and a machine carrying motors its task does not need is wasted payload. Off Earth, the winning robot is the lightest one that does exactly its job.

In other words, the off-world economy will belong to whoever has already mastered fully autonomous production on Earth. The habits we build in the factories of Ohio, Lyon and Winterthur this decade are the training ground for the foundries on the Moon the next.

If we lose the race for automated manufacturing here, we will not get a second chance up there. We will watch someone else build the first city beyond Earth, with its rules, its values and its flag. If we win it, the West's oldest instinct, to go further and build something that outlasts us, gets an entire sky to work with.

Why I care

I am French, American and Swiss. I do not experience the Atlantic as a border. I experience it as a hallway between rooms of the same house.

From San Francisco I learned that ambition is a renewable resource, and that a handful of people with conviction can bend the future. From France I inherited the idea that a nation can choose grand projects on purpose: high-speed rail, nuclear power, Ariane, Airbus. From Switzerland, where I studied at EPFL and now ETH Zurich, I learned something quieter and just as important: that a small, high-wage country with no oil and no coastline can still make the finest things in the world, if it treats precision as a way of life.

Those three lessons add up to one conviction. Prosperity is not a gift from history. It is manufactured, every day, by people who know how to make things and choose to keep doing it.

I do not want my generation to inherit a West that designs beautiful products for someone else to build, writes regulations for technologies it does not own, and visits its own former factories as museums. I want us to be the generation that made building feel inevitable again.

What I believe

  1. Making things is not a lower form of work. It is the foundation of every freedom we enjoy.
  2. Every step from the mine to the finished product can and must be automated, and the West can lead it.
  3. Robots should be built to their purpose, and intelligence should be shared. Optimization designs the bodies and the factory, learning runs inside them, and neither is enough alone.
  4. Physical intelligence is the next frontier of AI, and it is trained in factories. No factories, no frontier.
  5. Sovereignty is not autarky. It is having enough of your own capacity that your friends remain friends by choice.
  6. Western nations should compete hard and win together. Our rivalry is a race toward the same horizon.
  7. Hyperscale manufacturing lets us optimize form around the task, and it is how we scale beyond this planet.
  8. None of this is too late. But the window is measured in years, not decades.

We still have the best universities, the deepest capital, the frontier labs and the engineers who know how to make a part to the micron. We still have the ideals worth defending. What we have lost is the habit of building.

Habits can be rebuilt. Let's build millions of simple machines, each made for one job, all sharing one mind, and let them build everything else.

Sources