Imagine describing a gadget to an AI and watching a factory build it. No lengthy tooling process. No custom production line for every new design. Just instructions in—and a working machine out.
That is the extraordinary promise behind Atomic Machines, a California startup that emerged from six years of stealth on October 7, 2026. Its proposed leap is from AI that generates words and software to AI-guided manufacturing that produces physical devices.
The company calls its system the Matter Compiler. And although its vision sounds like something borrowed from science fiction, its first announced product is surprisingly practical: a power switch smaller than a fingertip.
A tiny switch with a big claim
Meet PrimeSwitch PS-150, an electromechanical relay designed for high-power applications such as AI data centers. According to Atomic Machines, the device measures just 9.5 millimeters across and 3 millimeters thick, carries 150 amps of continuous current and opens in 50 microseconds—just five hundredths of a millisecond.
Conventional mechanical relays conduct electricity efficiently through metal but generally switch more slowly. Semiconductor switches can react quickly, but their electrical losses produce heat. Atomic Machines says PrimeSwitch combines low-resistance metal conduction with unusually fast mechanical opening.
There is an important technical detail: the company describes using PrimeSwitch inside a hybrid breaker, where a separate solid-state component helps handle current during the moment the contacts open. A 50-microsecond opening time should not be mistaken for proof that the tiny switch, by itself, can safely interrupt every high-voltage fault.
Atomic Machines says it is supplying early-access customers with devices for evaluation. That is an intriguing milestone—but it is not the same as independently verified performance or widespread commercial deployment.
The Matter Compiler: a factory that learns?
Traditional manufacturing usually demands a specialized process for each device. Atomic Machines wants to replace much of that friction with a programmable system that builds, measures and learns. In the company’s description, each operation generates measurements that can inform corrections and subsequent attempts.
Founder Jeff Holden describes the long-term ambition as “prompt-to-product” manufacturing. In the imagined workflow, someone describes what a device should accomplish; software develops and checks a design, plans the fabrication steps and sends it for production. That is the “ChatGPT for hardware” analogy—not evidence that ChatGPT itself operates the machines.
The name Matter Compiler is also a deliberate nod to The Diamond Age, Neal Stephenson’s novel about machines that assemble physical objects from instructions. Reality, for now, is considerably more constrained: Atomic Machines focuses on micro-machines, not a household box that can manufacture anything you dream up.
Where the humans still matter
For all the automation language, the manufacturing operation is not yet a hands-off, lights-out factory. Reporting on the system indicates that engineers still transfer components between machines and help with assembly. Atomic Machines has demonstrated a first product family, not a public catalog of countless unrelated objects built on demand.
Questions remain about manufacturing yield, cost, scalability, reliability and independent testing. How many different device designs can the same setup reliably build? How quickly can it produce them? And can customers verify that the resulting parts meet their specifications?
A manufacturing revolution—or the first draft of one?
If the concept works broadly, it could eventually change how inventors, researchers and small businesses prototype specialized equipment. Tiny pumps, micro-robots, advanced cooling systems and medical tools are among the possibilities Atomic Machines points to. Those potential applications should not be confused with products already available.
For now, the more grounded story is compelling enough: a startup claims it can produce an unusually fast miniature power relay with a programmable, measurement-driven manufacturing system. The much bigger bet is that the same approach can eventually make physical invention feel more like writing software.
Apep Bar’s question: If inventing a machine becomes as simple as describing it, who will control the factories—and who gets to decide what they build?
Sources and further reading
- Atomic Machines: The Matter Compiler — the company’s stated vision and intended workflow
- Atomic Machines: Origin Story — founder’s description of the system and PrimeSwitch
- October 7, 2026, company announcement — published product specifications and early-access status
- The Next Web reporting — background, funding and questions about the technology
- Watch the original YouTube Short
Featured image is an illustrative electronics photograph, not a photograph of PrimeSwitch. Photo: Alexandre Debiève / Unsplash.
