Saturn Data

Saturn Data Inc. is a hardware startup in the San Francisco Bay Area building FPGA-accelerated servers optimized for high-memory, high-bandwidth workloads. Founded by Peter Schmidt-Nielsen, an MIT alumnus and former ML engineer at Redwood Research, the company leverages field-programmable gate arrays (FPGAs) as IO expanders to achieve extreme memory bandwidth and storage throughput in compact server form factors. Their core product is a 2U server featuring terabytes of DRAM, terabytes per second of memory bandwidth, over a petabyte of flash storage, and approximately a terabyte per second of bandwidth to said flash storage. The company targets search and vector database workloads as initial use cases, and operates on a partnership model where they port customer applications to their specialized hardware rather than selling general-purpose servers. The team consists of Peter Schmidt-Nielsen and Monica Valcourt, both MIT alumni with strong systems and hardware backgrounds. Peter co-authored the NeurIPS 2022 paper "Adversarial Training for High-Stakes Reliability" during his time at Redwood Research, demonstrating expertise at the intersection of AI safety and technical systems. Monica graduated from MIT in 2022 with a Computer Science degree and has experience in MEV trading, autonomous vehicles (Uber ATG), and semiconductor-related work. In 2024, Saturn Data received a $500,000 grant from Jaan Tallinn through the Survival and Flourishing Fund's Initiative Committee to support work on prototyping flexible hardware-enabled chip governors, or FlexHEGs. FlexHEGs are specialized hardware components that could be integrated into AI accelerators to enable trustworthy, privacy-preserving verification and enforcement of compliance policies for AI development -- such as limiting the scale of training runs or requiring safety evaluations. Saturn Data's deep FPGA expertise positions them to contribute to this critical AI governance infrastructure.