[AI Engineer Produces 1460 Solutions in 17 Hours, Potentially Saving Data Center Electricity $227,000 Annually] Beating AI News: Y Combinator-incubated startup Standard Physics has launched its AI engineer, Archimedes. Archimedes can design engineering solutions, run physical simulations, and continuously modify them based on the results to find better-performing designs. In a US data center, even with the cooling equipment running at full speed, the hottest server rack still reached 35.6°C. AI simulations revealed that some cool air wasn't being delivered correctly, and hot air exhausted from the servers was flowing back into the racks. The AI first tried adjusting the air vents and isolating hot air, then optimizing the cooling equipment's operation. The first round of energy-saving solutions seemed effective, but overlooked the actual capabilities of the old equipment. After reviewing and finding it unimplementable, the AI added equipment limitations and searched for new solutions. The entire process lasted 17 hours, trying 1460 designs. Two modification solutions were ultimately selected. According to the company's simulation calculations, a complete overhaul could reduce the total power consumption of the data center by 10.9%, lowering the temperature of the hottest server rack to 29.2°C, saving approximately $227,000 in electricity costs annually. Another low-cost solution achieves the same 79% energy savings with less than a quarter of the overhaul cost. The AI also tested 1,333 turbine blade designs in 2D simulations within 30 hours. The final design, while maintaining the same airflow steering requirements, reduced energy loss by 25.5% compared to the original blades, surpassing an optimized design published by a Belgian research team in 2018. The entire system is comprised of two parts: Archimedes is responsible for proposing designs, comparing results, and deciding on the next steps; Reynolds is responsible for running the simulations. It can convert problems described by engineers in natural language into OpenFOAM simulation tasks, automatically modeling, generating computational meshes, running calculations, and analyzing results. Standard Physics has open-sourced Reynolds, which is responsible for running the physical simulations, allowing anyone to use and modify the code for free. Archimedes, a fully-fledged AI engineer capable of independently designing solutions, repeatedly testing them, and finding the optimal solution, is not yet open-source; users need to contact the official team to use it. [Original Link]