Atomic Machines has introduced the PrimeSwitch, a new technology targeting 800V DC architectures for AI data center racks. The solution focuses on advanced fault protection and isolation to safeguard expensive compute hardware as power demands surge in hyperscale environments.
Overview: Atomic Machines Enters the 800V AI Power Arena
Atomic Machines has officially emerged from a six-year stealth period, introducing its first commercial product: the PrimeSwitch PS-150. Unlike the company’s broader, futuristic Matter Compiler pitch, the PrimeSwitch is a highly specific electromechanical relay designed to solve a critical engineering challenge inside next-generation AI data centers. The device targets the protection of 800-volt DC power architectures without dissipating excessive energy as heat. Serving as the mechanical half of a hybrid DC protection system, the PrimeSwitch represents a concrete engineering milestone for Atomic Machines and its AI-native manufacturing platform.
Technical Details: Bridging the Gap Between Mechanical and Solid-State
The PrimeSwitch PS-150 boasts a set of specifications that make it uniquely suited for high-density power environments. The relay can carry 150 amps continuously with an exceptionally low on-resistance of 200 micro-ohms. At full current, this resistance yields a voltage drop of approximately 30 millivolts, resulting in a resistive loss of roughly 4.5 watts. Furthermore, the device is bistable, meaning it consumes zero holding power. When a fault occurs, it opens to provide 1,500 volts of galvanic isolation in just 50 microseconds. Rather than acting as a standalone breaker, it functions as the fast mechanical contact inside a hybrid system, where a parallel semiconductor temporarily manages the fault current during the brief microseconds required for the mechanical contact to fully open.
Market Context: The Shift to 800V DC in Hyperscale Racks
The introduction of the PrimeSwitch aligns with a massive architectural shift in hyperscale AI data centers. As AI racks become increasingly power-dense, the industry is transitioning toward 800V DC distribution. Operating at higher voltages allows for lower current at the same power level, which significantly reduces copper requirements and conduction losses. This transition is already underway, with major industry players like Nvidia collaborating with Foxconn to implement 800V DC power architecture in the new K-1 AI data center in Kaohsiung, Taiwan, moving away from the traditional 54V standard. While other semiconductor companies are focusing on collapsing this 800V power directly to low-voltage GPU rails, Atomic Machines is targeting the critical upstream phase: fault protection before the power reaches expensive compute hardware.
Industry Impact: Redefining Fault Protection and Isolation
Protecting high-current DC circuits has historically required a compromise between speed and efficiency. Traditional mechanical contactors offer excellent metal-to-metal conduction with extremely low resistance and provide true physical separation when open. However, they are slow, often taking milliseconds to separate, which is inadequate for rapidly rising DC faults. Conversely, solid-state switches react in microseconds and avoid mechanical arcing, but they conduct through silicon or wide-bandgap materials, continuously dissipating power as heat and lacking true physical galvanic isolation. By combining these approaches, Atomic Machines aims to deliver the low-resistance efficiency of a mechanical relay with the microsecond-level fault response of a semiconductor, fundamentally changing how DC protection is architected.
Implications for AI Data Centers: Thermal and Efficiency Gains
In the context of AI data centers, the 4.5-watt resistive loss figure is highly significant. A switch carrying hundreds of amps operates in an environment where every additional milliohm translates directly into heat, increased cooling loads, and lost electrical efficiency. By minimizing the energy spent simply moving power through protection hardware, the PrimeSwitch allows facilities to maintain higher compute densities. While it sits farther upstream than recent innovations like Infineon’s 27 kW power supplies for AI server racks, both technologies highlight a critical reality: as rack-level power climbs, components previously treated as mere supporting hardware now directly dictate overall compute density and thermal limits.
Future Outlook: Validation and the Matter Compiler Vision
Despite its impressive specifications, the PrimeSwitch is not a complete circuit breaker in a single package. A practical hybrid breaker still requires fault detection circuitry, a parallel semiconductor path, energy absorbers, and system-level isolation components. Furthermore, technical analyses note that Atomic Machines has not yet publicly specified crucial production metrics, such as actuation-drive requirements, temperature derating, and endurance under repeated fault-opening events. For Atomic Machines, the PrimeSwitch is a vital proof-of-concept. If early-access customers validate its electrical specifications and reliability in real-world data center conditions, it will prove that the company's Matter Compiler platform can successfully manufacture functional micro-machines at scale, turning digital code into critical physical infrastructure.