We apply high-fidelity conjugate CFD inside multi-objective Bayesian optimization to design thermal hardware for AI compute — GPU cold plates, heat sinks, and data-center cooling. Each study pairs validated simulation with an optimizer that reaches the best design in far fewer runs, and reports the full performance trade-off rather than a single operating point.
Conjugate CFD and Bayesian optimization over microchannel geometry: the thermal-resistance vs. pumping-cost trade-off, a design roughly 15% cooler within budget, and the warm-water free-cooling power case.
Conjugate (solid–fluid) heat-transfer simulation in OpenFOAM®, validated against published experiments and convergence-gated on every run; multi-objective Bayesian optimization to map the full design trade-off; and physics-informed surrogates that cut the number of expensive simulations roughly in half.
Given a component and operating specification, we produce the optimized design and the trade-off analysis for the actual part.
OpenFOAM® is a registered trademark of OpenCFD Ltd. This work is not endorsed by OpenCFD Ltd.
© 2026 Toaiz LLC · toaiz.ai