A Bay Area start-up called Atomic Machines has come out of six years in stealth with $250 million in funding and a bold pitch: a factory that builds tiny working machines directly from code. The company calls its system the Matter Compiler, and it announced the launch on October 7, 2026.
Its founder and CEO is Jeff Holden, who joined Amazon as its tenth engineer, led the development of Amazon Prime and later became Uber’s first chief product officer.
What the Matter Compiler does
The name comes from a machine in Neal Stephenson’s novel The Diamond Age. According to the company, the code alone decides which device gets built. There are no moulds, masks or other fixed tooling, so changing a design is a software change rather than a factory change.
Investor Construct Capital describes the system as a set of automated “Nodes”, each running one type of process. Lasers machine parts, fluids are deposited in picolitre drops, electrochemical steps plate and polish, and robots move and assemble the pieces. The materials include metals, polymers, glasses and ceramics.
The company says it can make machines with moving parts, several materials and features down to single-digit microns. The system measures and corrects each step as it goes, and its AI runs its own experiments on the hardware. Holden compares it to the ribosome, the part of a cell that builds proteins from genetic code.
Why does this matter? According to Atomic Machines, conventional manufacturing hits its limits at a few tens of microns. Below that, chipmaking tools take over, and each new micro-device (known as MEMS) needs years of custom process work.
The first product: a fast switch for AI data centres
The first device is the PrimeSwitch PS-150, a relay meant for AI data centres that are moving to 800-volt direct current power. At those voltages, the company says, a fault can release destructive energy in microseconds, while mechanical protection takes milliseconds to open.
- It opens in 50 microseconds, which the company says is about 1,000 times faster than a conventional contactor.
- It carries 150 amps of continuous current.
- The package is 9.5 mm across and 3 mm thick.
PrimeSwitch is shipping to unnamed early-access customers for evaluation. Its first public showing is at the Open Compute Project Global Summit in San Jose, from October 12 to 15.
Who is backing it
Investors include OneIM, Gigafund, XTX Ventures, KAS Venture Partners, Valor Equity Partners, Sozo Ventures, Construct Capital and the Regents of the University of California. According to The Next Web, the company did not share a valuation or how the $250 million was split across rounds.
“If vibe coding is how software gets made now, this is the start of vibe manufacturing,” said Luke Nosek, managing partner at Gigafund.
Two sides of the story
The optimistic view is easy to see. If small, complex machines can go from design to device in hours instead of years, that could open doors in medicine, electronics and cooling. The company lists gears the size of a grain of sand, robots small enough to work inside the body and cooling built into AI chips as targets.
The cautious view is just as fair. Almost all of the performance claims so far come from the company and its investors. Customers aren’t named, the switch hasn’t been independently tested in public yet, and turning one product into a broad “build anything” platform is a much harder step. Tiny medical robots in particular would face strict safety testing and regulatory approval before reaching patients.
Why this caught my attention
As a writer, I think the most interesting part isn’t the $250 million. It’s that the first product is a humble power switch, not a flashy robot. Starting with a real, sellable part for a booming market like AI data centres feels sensible to me. Still, “vibe manufacturing” is a catchy phrase, and I’d like to see independent tests and named customers before calling this a manufacturing revolution. For India, which is building its own electronics and data centre base, it’s a space worth watching.
For more context, read about L&T’s Nvidia AI data centre order, the $95 billion power conversion market, and why manufacturing leaders underfund AI’s foundations.