Perstorp's Shubha Mitra Basu presents a case for replacing water-glycol direct-to-chip coolants with PFAS-free, biodegradable synthetic ester dielectric fluids in data centers, addressing heat as the central bottleneck limiting AI infrastructure performance and sustainability. The talk covers data center energy and carbon metrics, Perstorp's SYNMERSE product line, a collaborative performance demonstration with Intel SuperFluid and Inventec system design, and the fluid's material compatibility validated to OCP Immersion Requirements. The session concludes with sustainability data on AI energy growth and the carbon footprint advantage of consciously-sourced synthetic esters.

Key Takeaways

• US data centers consumed 176 TWh in 2023 (4.4% of US electricity) and are projected to hit 12% of US electricity by 2028; cooling alone accounts for ~50% of that energy draw.
• Switching to dielectric synthetic ester fluids eliminates electrical damage on leakage, removes the need for fluid corrosion treatment, and reduces system mechanical complexity versus water-glycol.
• Perstorp's SYNMERSE DC-L (5.6 cSt, flash point 180°C), DC (16 cSt, 247°C), and DC-R (29 cSt, 263°C) are all PFAS-free and readily biodegradable with volume resistivity >1×10¹² Ω·cm.
• Combined Perstorp DTC fluid + Intel SuperFluid air-infusion technology (Inventec system) achieved 930 W TDP at 0.5 LPM and 1,100 W at 0.7 LPM, reducing thermal resistance by 8.8–11.4% versus non-SuperFluid baseline.
• Perstorp tested fluids across more than 200 material types (seals, metals, plastics, cables, hoses) in cooperation with OEMs/ODMs, validating against OCP Immersion Requirements Rev 2.10 (2023-08-18).
• By conscious raw material sourcing, the Synmerse DC product carbon footprint can be reduced by a further 92% beyond the already-low baseline PCF.
• AI workload growth has driven a 150% rise in tech-sector CO2 emissions from 2020–2023; GPT-5 queries use ~18 Wh each, and 2.5 billion daily queries could require 45 GWh/day — equivalent to powering 1.5 million US households.