Size often matters in technology, and the Matter Compiler is no exception.

Its promise lies in the extraordinarily tiny machines it can produce, potentially opening the door to everything from microscopic medical robots to miniature motors, pumps and cooling systems that conventional manufacturing has struggled to create.

Developed by Atomic Machines, a San Francisco Bay Area technology company, the Matter Compiler represents an attempt to change how physical objects are manufactured. Rather than relying on specialized equipment, molds and complicated production processes for each new invention, the system uses AI and digital instructions to produce working machines with moving parts from computer-generated designs.

Imagine designing a tiny mechanical device on a computer, submitting those instructions to a manufacturing system and receiving a functioning machine without having to build an entirely new production line. That is what Atomic Machines is pursuing: a future in which making a physical object could become almost as programmable as creating software.

After six years of working in secrecy, the company emerged on Oct. 7, announcing that it had raised $250 million from investors to develop the technology. Founded by Jeff Holden, a former Amazon executive who helped create Amazon Prime and later served as Uber’s chief product officer, Atomic Machines has offices in Emeryville and Santa Clara, California.

Manufacturers have struggled for decades with a fundamental problem when attempting to shrink mechanical devices to microscopic dimensions. Traditional methods, such as cutting, molding and assembling individual components, become increasingly difficult as parts become smaller. Semiconductor factories can produce intricate electronic circuits, but their equipment is largely designed to create flat structures using a limited selection of materials.

Engineers can draw up blueprints for microscopic machines with gears, motors, valves and other moving parts, but producing them often requires years of specialized engineering, assuming they can be made at all.

Atomic Machines believes the Matter Compiler is the answer.

The system combines several manufacturing processes into a single, computer-controlled platform capable of working with different materials and producing complex, three-dimensional structures. Unlike a conventional 3D printer, which typically creates an object by depositing or solidifying material in successive layers, the Matter Compiler is designed to manufacture complete mechanical devices, including their internal components and moving parts.

At the heart of the system is AI, which helps determine not only what a machine should look like but also how to make it. A designer supplies digital specifications describing the intended device, and the software develops a manufacturing plan, determining the sequence of operations needed to turn the design into a physical object.

Precision measuring instruments examine the work and provide feedback to the AI. If something deviates from the intended design, the system can adjust its operations, using information gathered during manufacturing to improve subsequent attempts.

The Matter Compiler is designed to function as a factory that can evaluate its own work and learn from the results. Each new design would not necessarily require engineers to create an entirely new manufacturing process, potentially reducing development time from years to weeks or even days, according to the company.

Atomic Machines wants to simplify the process even further, allowing someone to describe a desired machine in ordinary language and have AI design and manufacture it. The company calls that goal “prompt-to-product.”

The potential applications extend beyond conventional electronics.

In medicine, the technology could eventually enable microscopic robots capable of traveling through parts of the human body that conventional surgical instruments cannot easily reach. Miniature pumps could move tiny quantities of medication or biological fluids, while small diagnostic devices could perform sophisticated laboratory tests without requiring bulky equipment.

In computing, microscopic cooling systems could be integrated directly into processor packages, potentially helping dissipate the enormous amounts of heat generated by powerful AI chips. Other possibilities include miniature motors, precision sensors and mechanical components that could make everything from industrial robots to spacecraft smaller and more efficient.

The company is beginning with a much narrower application intended to demonstrate that its manufacturing approach can produce something commercially useful. That first product is PrimeSwitch, an electrical relay designed to control the flow of power.

Relays operate like switches, opening and closing electrical connections to control or interrupt current. Conventional mechanical relays are efficient at conducting electricity but can be relatively slow to disconnect. Electronic switches can respond much faster, although they may generate additional heat and waste energy during operation.

PrimeSwitch attempts to combine the advantages of both technologies. According to Atomic Machines, it can open an electrical connection in just 50 microseconds, or 50 millionths of a second, roughly 1,000 times faster than a conventional mechanical contactor. Its tiny moving components travel extremely short distances, allowing the device to operate at speeds that larger mechanical switches cannot easily achieve.

The device measures just 9.5 millimeters across and 3 millimeters thick, making it small enough to fit on a fingertip. Yet it is designed to handle substantial electrical loads, including those associated with increasingly power-hungry AI data centers.

As those facilities move toward higher-voltage electrical systems, rapidly disconnecting power during an equipment failure becomes increasingly important. PrimeSwitch could help protect sensitive equipment while reducing the energy losses associated with conventional switching arrangements.

Atomic Machines says the device is already being shipped to early-access customers for evaluation. The company plans to showcase PrimeSwitch at the Open Compute Project Global Summit in San Jose, California, Oct. 12-15.

For Holden, the larger opportunity lies in changing the economics of invention.

“3D printing made shapes programmable: any geometry, straight from a file. We’re making machines programmable: many materials, moving parts, sealed and wired, straight from a file.”