I started watching Star Trek when I was five, and it never let go. What Gene Roddenberry gave me, through Kirk, Spock, McCoy, and the rest, was a conviction: that exploration in pursuit of knowledge is a powerful purpose in and of itself. And that new knowledge, and the technology it enables, lets us transform zero-sum situations into positive-sum ones and prepare for problems we don’t even know we have yet.
Years of reading and thinking about how civilizations actually advance eventually hardened that conviction into a thesis, and the thesis into a company.
The thesis is this: civilizational progress rests on three axes of innovation: the command of energy, the command of information, and the command of matter. Knowledge itself is not a fourth axis; it is the traveler on the three. A civilization’s position on each axis is nothing other than its inventory of embodied knowledge of that type.
Looking hard at all three axes, two things became clear.
The first is that we are in the early innings everywhere. Measured against what physics permits, humanity’s total energy harvest has roughly five-hundred-fold headroom before we even reach planetary scale, and ten further orders of magnitude beyond that at solar system scale. Our total computational throughput sits dozens of orders of magnitude below the theoretical limits of ordinary matter; by one well-known calculation11 Seth Lloyd, ‘Ultimate physical limits to computation,’ Nature 406, 1047–1054 (2000). Lloyd’s limit implies ~10²⁵ bits per milligram. Hilbert & López (Science, 2011) estimated humanity’s stored information at ~10²¹ optimally compressed bits as of 2007; the installed storage base has since grown by roughly two orders of magnitude, and current estimates run ~10²³ to 10²⁴ bits depending on what counts as stored., everything humanity has ever stored would fit comfortably inside a milligram of matter organized to its physical maximum, a single grain of sand. The vast majority of our learning and invention is in front of us, on every axis.
The second is that the axes are not peers: the command of matter is the one the other two run on, and the one where they are cashed in.
“Cashed in” means this: a joule is fungible and a bit is abstract. Neither touches the world except through an arrangement of atoms. A civilization’s energy is worthless until it turns a shaft or drives a synthesis; its information is inert until it steers a machine or edits a genome. Every impact either axis has ever had was delivered as a change in the arrangement of matter, because a change in the arrangement of matter is what an impact is.
And matter is the only axis that hosts itself. Every instrument of the command of information has been a matter artifact: the clay tablet, movable type, the telegraph’s copper, the processor’s doped silicon. Every converter in the command of energy has been a matter artifact: the boiler plate, the turbine blade, the photovoltaic wafer. But the instruments of the command of matter are themselves matter: the furnace, the lathe, the lithography line. One axis runs on itself; the other two run on it.
History repeats the point at every regime change. The Scientific Revolution rode on shaped glass: telescope, microscope, prism. This is centuries of furnace craft delivering science its sensorium. James Watt had the separate condenser years before he had an engine; what stood between them was John Wilkinson’s boring machine, a matter breakthrough that unlocked an energy regime. And in December 1947, a triumph of material command in the forms of purified germanium, controlled impurities, and a physically formed junction, launched the Information Age. Every time civilization has jumped to a new regime, the rate-limiting step was the matter step.
So what is the state of humanity’s command of matter, measured against what’s possible?
Here the most illuminating benchmark comes from biology, because nature has already built the machine we should be measuring ourselves against: the ribosome. It reads a string of digital symbols and, from that string alone, constructs a working molecular machine. A protein 150 amino acids long, drawn from an alphabet of 20, gives 20¹⁵⁰ expressible structures, from one unchanging machine. Across four billion years, evolution has sampled perhaps 10¹⁶ unique proteins through this single system: pumps, motors, hinges, sensors, catalysts. And after four billion years, the marginal cost of the next design has never risen: one more machine costs one more message, plus feedstock.
Now compare how humanity makes things. In every human manufacturing system, from the machine shop to the leading-edge fab, the design lives in the hardware. Every new artifact requires a new physical embodiment of its design: a mold, a mask, a die, a fixture, a retooled line. Our catalog of things grows by accumulating machinery, one paid-for entry at a time. In the ribosome’s system, the design lives in the data: the machine is fixed, and new artifacts are new messages.
To state the distinction plainly, an additive repertoire is a list; a combinatorial repertoire is a language. A list grows by appending entries, each purchased separately. A language’s sentences were never enumerated and never had to be: they exist the moment the grammar does. That is why the ribosome commands 10¹⁶ structures having pre-built none of them.
You might object that humanity does possess combinatorial systems, such as a 3D printer over shapes or an e-beam writer over patterns. True, but look at the unit of those languages: a passive shape. Biology’s unit is the working machine. And by that standard, the honest assessment of our species’ command of matter is stark: no human system has ever had a combinatorial repertoire whose elements are functioning devices. Everything we make, we make from the list.
The gap is widest where the territory is largest. Our macroscale catalog (forging, casting, molding, machining, three centuries of accumulated industry) does not shrink gracefully as you descend the length scale. It disintegrates, process by process, each at its own physical wall, generally by the low tens of microns. Below that line, essentially one industrial capability survives: semiconductor fabrication. It is a magnificent capability, and it was built to make exactly one kind of machine, namely information machines in silicon. Micro-machines that move, pump, heat, focus, and sense (the mechanical, fluidic, thermal, optical, and electromagnetic domains) remain effectively unmanufacturable. The Information Age industrialized one province of the microscale and we have been living inside it ever since.
This is why I founded Atomic Machines, and why our mission is on-demand universal command of matter.
We have spent six years in stealth building the Matter Compiler: the first manufacturing system in history with the same architecture as biology’s: a combinatorial constructor whose unit of output is the working machine. The name is a tip of the hat to Neal Stephenson, who imagined the endpoint beautifully in The Diamond Age; we are building the real, bounded, near-term beginning of it, starting at the microscale.
The Matter Compiler is all-digital, with zero hard tooling: no molds, no masks, no fixtures. It is driven entirely by symbolic instructions, which means the marginal cost of a new device design is the cost of writing it down. Its operations are composable primitives that take design intent as input, rather than recipes rediscovered by trial and error for every new part. And every operation it performs is measured, so that errors are caught and corrected instead of compounding, the critical property that lets construction programs run deep. Those three properties are what make a repertoire a language rather than a list. The ribosome has them. Now a machine built by human beings does too.
The first thing we have said in that language is PrimeSwitch, which we are also unveiling today. We chose it six years ago from first principles. Power density was going to keep climbing, and the two ways of protecting a power bus each fail half the job: semiconductors open in microseconds but pay for that speed with heat the whole time they conduct, while electromechanical relays conduct efficiently but take milliseconds to open. Electrification exposed that gap first; the move to 800-volt DC in AI data centers has made it acute. PrimeSwitch PS-150 is the device that closes it: a relay that holds its state with zero power, is rated for 150 amps of continuous current, and opens in 50 microseconds. It is a multi-material micro-machine that moves, in the territory I said above was unmanufacturable. Semiconductor fabs cannot process its materials or assemble its moving parts, and no conventional tool can reach its feature sizes. It is the first device that exists because the language exists.
Because the Matter Compiler’s language is symbolic, it is also AI’s native medium: the same data type generative models read and write. Our destination is what we call prompt-to-product manufacturing: an experience like working with a modern AI coding agent, except that some of the output tokens are physical.
PrimeSwitch matters. But the machine matters more, because a language is defined by everything it can say, not by its first sentence. The list regime took humanity three hundred years to build. The language regime starts today, and we are in the early innings of everything.
- Seth Lloyd, ‘Ultimate physical limits to computation,’ Nature 406, 1047–1054 (2000). Lloyd’s limit implies ~10²⁵ bits per milligram. Hilbert & López (Science, 2011) estimated humanity’s stored information at ~10²¹ optimally compressed bits as of 2007; the installed storage base has since grown by roughly two orders of magnitude, and current estimates run ~10²³ to 10²⁴ bits depending on what counts as stored.