Atomic Machines Raises $250M for Matter Compiler

At 800 volts, the difference between 50 microseconds and a few milliseconds is enough time for fault energy to become damage. That gap sits inside the power architecture now being designed for AI data centers, where rack density is rising faster than the mechanical protection equipment built for an earlier electrical era can react.

Atomic Machines emerged from six years in stealth on October 7, 2026 with a proposed answer: PrimeSwitch PS-150, a micro-scale electromechanical relay designed to open in 50 microseconds while carrying current through metal contacts. The company also disclosed that it has raised $250M to date from investors including Gigafund, OneIM, XTX Ventures, KAS Venture Partners, Valor Equity Partners, Tru Arrow Partners, Construct Capital, Sozo Ventures, and the Regents of the University of California.

The capital accounting matters. Atomic Machines did not announce a named $250M round, disclose a lead investor, or provide a valuation. Its official launch announcement describes $250M as cumulative funding raised across the company's life, financing an unusually long attempt to turn manufacturing knowledge from product-specific hardware into reusable code.

A Factory That Treats a Product Like Software

Conventional manufacturing stores much of a product's design in physical commitments. A mold, mask, fixture, process recipe, assembly line, and inspection plan remember what the factory knows how to make. Change the product and enough of that memory has to be bought again, which is why high variety and high volume usually pull production in opposite directions.

Atomic Machines is building what it calls the Matter Compiler, an AI-native manufacturing system that uses symbolic instructions to produce functional, multi-material micro-machines. The company says the system measures each operation, corrects errors in real time, and lets AI test what simulation proposes on the same hardware that will make the device. Its goal is a manufacturing language in which a new design becomes a new instruction set rather than a new factory process.

That is a larger claim than faster prototyping. Semiconductor fabs can pattern extraordinary electronic structures, but their materials and flat geometries were optimized for circuits. Traditional machining, casting, and molding lose reach as features shrink toward the low tens of microns. Atomic Machines is targeting the territory between them: moving parts, pumps, relays, gears, thermal systems, optical devices, and other three-dimensional machines with features measured in single-digit microns.

PrimeSwitch Makes the Thesis Testable

The first product is PrimeSwitch PS-150, a bistable electromechanical relay built for high-voltage power systems. Atomic Machines lists a 150-amp continuous-current rating, 50-microsecond opening time, 200-micro-ohm on-resistance, 1,500 volts of galvanic isolation, zero holding power, and a 9.5-millimeter by 3-millimeter surface-mount package.

Those specifications aim at a real engineering trade-off. Mechanical contactors conduct efficiently and create a physical isolation gap, but they open in milliseconds and can arc. Solid-state switches react in microseconds, yet conduct through silicon that produces heat and cannot create the same galvanic isolation. Atomic Machines pairs PrimeSwitch with a small solid-state bypass so silicon handles the instant of interruption while the metal contact carries and isolates during normal operation.

The company says PrimeSwitch is shipping to unnamed early-access customers for evaluation and will make its public debut at the Open Compute Project Global Summit in San Jose from October 12 through October 15. The specifications and performance comparisons remain company-reported until customers or independent testing provide field evidence, but the product gives the Matter Compiler thesis somewhere concrete to fail or succeed.

The People Behind the Machine

Jeff Holden, Atomic Machines' founder and CEO, joined Amazon as an early engineer, helped build Amazon Prime, and later became Uber's first chief product officer. In his origin story, he frames the company around a long-range belief that progress in energy and information ultimately depends on the ability to arrange matter.

CTO Klaus Zietlow brings a more immediate bridge to commercial machinery. Atomic Machines credits Zietlow with system-architecture work on Verb Surgical's robot and contributions at Intuitive Surgical, Agilent, and AccuVein. The official leadership team also spans device engineering, simulation, manufacturing, platform software, finance, people, and go-to-market functions, reflecting how little of this problem belongs to one discipline.

Capital for an Industrial Proof

The $250M total funded six years of development before the company put a product in public. That patience is understandable for a business building both a manufacturing platform and the first device meant to prove it, but it also raises the standard for what comes next. A programmable fab has to establish repeatability, yield, throughput, reliability, and economics, not just produce a compelling prototype.

The physical expansion is already visible. In November 2025, California GO-Biz reported that Atomic Machines planned to invest $156.3M in MEMS manufacturing facilities across Santa Clara, Emeryville, and the broader East Bay, with 305 proposed new jobs. The company's careers page listed 21 openings across AI, engineering, manufacturing, security, finance, people, and systems when checked on October 8, 2026.

That expansion is separate from the disclosed funding total and does not establish how the company allocated its capital. It does show the scale of the operating commitment: Atomic Machines is moving from a closed research program toward production, customer evaluation, and a workforce capable of supporting both the factory and the products coming out of it.

What This Changes for Advanced Manufacturing

If Atomic Machines can make one capital base serve many device designs, the manufacturing economics become more interesting than PrimeSwitch alone. Product teams could test hardware without waiting years for a custom process, smaller markets could support specialized devices, and factories could add variety without rebuilding the production system around each SKU.

The opportunity reaches beyond data-center power into medical devices, thermal management, fluidics, sensing, robotics, and aerospace. The same breadth creates the risk. A general manufacturing system has to master materials, interfaces, metrology, software, and quality control across device classes that established industries normally separate into specialized processes.

Atomic Machines has raised enough capital to move that argument out of stealth. PrimeSwitch now carries the first public burden of proof, and every early-access evaluation will test whether a manufacturing language can survive the far less forgiving grammar of heat, current, materials, yield, and time.