Physical Superintelligence, an AI startup, has closed a $58 million seed round led by Breakthrough Energy Ventures to deploy machine learning agents designed to optimize data center cooling, power and electrical systems before facilities are built.
The company's platform, called Emmy, uses AI agents paired with built-in physics verifiers to catch outputs that are confident but wrong, according to the announcement. The startup says its initial focus is on terrestrial and orbital data center optimization, with commercial projects already in motion.
Breakthrough Energy Ventures, the investment vehicle founded by Bill Gates, led the round. The venture firm has backed companies including Commonwealth Fusion Systems and Twelve, both focused on industrial applications of emerging technology. Emmy embeds physics constraints into AI reasoning to reduce hallucination and improve output reliability in technical domains.
Data center operators face rising pressure to manage energy consumption as AI workloads grow. A single large language model training run can consume millions of kilowatt-hours, and cooling represents 30 to 40 percent of operational expenses at many facilities. Optimizing these systems before construction, rather than retrofitting them, eliminates the cost of redesign.
Emmy's approach differs from conventional simulation tools, which typically require human engineers to set up and interpret results. The startup claims its agents can autonomously explore design spaces and propose solutions that account for interdependencies between mechanical, electrical and thermal systems. The physics verifiers are designed to reject outputs that violate conservation laws or other physical constraints, even if the model's confidence score is high.
The data center optimization market includes competitors ranging from established firms like Schneider Electric to specialized startups. Physical Superintelligence's entry into the space comes as large cloud providers including OpenAI, Google and Meta are investing heavily in their own chip and facility design capabilities. The startup's focus on pre-construction optimization targets a narrower problem: helping operators and builders reduce energy spend and capital cost before equipment installation.
Physics verifiers embedded into commercial engineering products remain relatively rare. Physical Superintelligence is backing this claim with paying customers already using the platform.
The startup will need to demonstrate that its AI-driven optimization produces measurable savings once facilities are operational. If Emmy's recommendations yield lower energy consumption or capital costs than conventional design methods, the startup would have a direct metric to market to other operators. The number that decides its viability is whether real-world data center performance matches the platform's pre-build projections.
