Space-Station Tech Pivots to Cool AI Data Centers

2026-05-20 · Source: Big Data & AI News - EE Times · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure · Depth: Intermediate, medium

Summary

Mikros Technologies and Carbice Corporation are adapting technology originally developed for the International Space Station to address critical heat dissipation challenges in AI data centers. Mikros, acquired by Jabil in 2025, specializes in microchannel liquid cooling, partnering with Broadcom for its 5-kW XPU custom accelerators and Marvell for co-packaged optics chips. Their MikroMatrix design precisely distributes coolant to match GPU power maps, ensuring even cooling and enabling higher rack densities. Carbice Corporation provides carbon nanotube Thermal Interface Materials (TIMs) that mechanically adjust to chip curvature changes, preventing performance throttling. Carbice's material increases surface area for convection by over 10,000 and has been tested on Nvidia chips at Georgia Tech for three years. These innovations are crucial given increasing chip heat emission, which caused a May 7 Amazon data center shutdown, and data centers' projected 12% of total U.S. electricity consumption by 2030. Liquid cooling can cut over a third of energy consumption from air-cooled data centers and save over \$1 million per rack lifetime.

Key takeaway

For AI Hardware Engineers and MLOps teams designing or upgrading high-density data centers, you must prioritize advanced thermal management. Integrating microchannel liquid cooling, like Mikros's MikroMatrix, and adaptive carbon nanotube TIMs, such as Carbice's, is crucial to prevent costly throttling events and maximize full ASIC performance. This approach ensures system stability, reduces energy consumption by over a third, and supports the 5-kW XPU demands of future AI compute.

Key insights

Space-station derived liquid cooling and carbon nanotube TIMs are critical for managing extreme heat in next-gen AI data centers.

Principles

• Chip heat emission drives demand for advanced cooling.
• Liquid cooling significantly reduces energy consumption.
• TIMs must adapt to chip curvature changes.

Method

Mikros's MikroMatrix design uses a matrix array of microchannels perpendicular to the chip surface to distribute coolant, matching GPU power maps for even, efficient cooling. Carbice's carbon nanotube TIMs mechanically adjust to chip curvature.

In practice

• Implement microchannel liquid cooling for 5-kW XPUs.
• Use carbon nanotube TIMs for stable chip-to-cold plate contact.
• Design for liquid cooling in next-gen AI chips.

Topics

• AI Data Centers
• Liquid Cooling
• Thermal Interface Materials
• Carbon Nanotubes
• Microchannel Cooling
• Broadcom XPUs
• Marvell Co-packaged Optics

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Hardware Engineer, AI Architect, MLOps Engineer

Related on AIssential

• AI, energy and infrastructure: The future of data centre infrastructure in a high-demand world — Balancing escalating AI-driven compute demand with stringent sustainability goals requires advanced cooling, renewable energy, and intelligent optimization.
• This new chip survives 1300°F (700°C) and could change AI forever — A new memristor design withstands 700°C, enabling extreme environment electronics and efficient AI computation.
• Israeli AI cooling startup ZutaCore raises $100 million Series C at estimated $600 million valuation - CTech — ZutaCore's HyperCool offers a waterless, two-phase direct-to-chip liquid cooling solution for high-density AI data centers.
• Liquid-cooled AI systems expose the limits of traditional storage architecture — Hybrid air/liquid cooling for AI infrastructure is inefficient and creates thermal liabilities for storage.
• This Will Power Everything — Optical interconnects are critical for scaling AI data centers beyond copper's physical and thermal limitations.
• IBM, Synopsys Move Toward 1.4-nm Node With Heat-Modeling Tech — A new ML-powered thermal modeling tool accelerates nanoscale chip design, enabling 1.4-nm nodes and beyond.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Big Data & AI News - EE Times.