Jeff Holden Returns with $250 Million and Matter Compiler
Jeff Holden, an early Amazon engineer and builder of Prime, returns after a six-year absence under the name Atomic Machines, bringing a total of $250 million in funding and a factory to launch a system called Matter Compiler.
The system is described as building micro-machines with moving parts, multiple materials, and three-dimensional structures directly from code, with feature sizes down to single-digit micrometers; different products can be produced on the same facility by simply changing the code, without needing to reconfigure dedicated processes or tools for each product.
Holden named the system Matter Compiler, inspired by the concept of a matter compiler from Neal Stephenson's 1995 novel "The Diamond Age"; the company claims it measures and corrects every step during the building process and feeds the learned data back into the next design, creating a data flywheel on the manufacturing side.
The first product is the PrimeSwitch PS-150, a 150-amp relay for 800-volt DC power supplies in AI data centers, with an opening time of about 50 microseconds, approximately 1000 times faster than traditional contactors, a conduction resistance of 200 micro-ohms, and a package size of 9.5 mm by 3 mm, which has been sampled to early customers.
Traditional chip factories excel at processing flat silicon wafers for information, but must leave dedicated production lines when faced with moving parts, multiple materials, and three-dimensional structures; new equipment often requires years and tens of millions of dollars from blueprint to mass production; the "one product, one process" approach has long hindered reusable Moore's law iterations for micro-machines.
Holden's approach is to introduce the versatility of chip factories into the manufacturing of machines with moving parts, allowing design changes to primarily occur in software rather than through reconfiguration of production lines; funding and attention are directed towards a universal micro-manufacturing system that can eliminate the barriers of dedicated processes, benefiting from AI infrastructure that can rapidly iterate hardware, while traditional micro-mechanical paths are pressured by reliance on dedicated molds and long-cycle processes.
Investors in the company include OneIM, Gigafund, XTX Ventures, and Valor Equity Partners, and the PrimeSwitch will be publicly showcased at the Open Compute Project summit.
Source: Public Information
ABAB AI Insight
Holden transitioned from being Amazon's 10th engineer and contributing to Prime's development to becoming Uber's Chief Product Officer and founding autonomous driving and AI labs, before spending six years in stealth developing a system that directly converts code into micro-machines with moving parts; the path shifts from software products to the generalization of hardware manufacturing.
On the capital front, the company has raised $250 million, directing resources towards an AI-native, fully digital manufacturing platform, with the first product targeting high-speed relays needed for 800-volt power supplies in AI data centers; the motivation is to eliminate the barriers of "one product, one process," enabling micro-mechanics to achieve reusable iterative capabilities similar to chips.
Similar cases include semiconductor manufacturing achieving long-term scaling through generalized processes, and 3D printing digitizing geometric shapes, but failing to simultaneously handle multiple materials and moving parts; the current phase is an attempt to extend generality from information processing to movable micro-machines.
Essentially, this represents a restructuring of the industrial chain and the generalization of processes: shifting manufacturing from dedicated molds and long-cycle processes to a code-driven, real-time measurement and correction system, moving value from one-time process development to a cumulative learning manufacturing flywheel; the sci-fi concept of a matter compiler first materializes in the least glamorous power protection devices, rather than in consumer-facing machines.
ABAB News · Cognitive Laws
Moore's Law belongs to industries where design changes do not require changing production lines, not to those that can only use dedicated processes.
The matter compiler in sci-fi often first appears as a switch in a data center, rather than a universal machine in a home.
Once manufacturing has its own data flywheel, the speed of iteration begins to shift from process cycles to software cycles.