DigestAI news desk
Hardware & Compute updated 1 min read

Single-phase direct liquid cooling validated for next decade of ultra-dense AI compute

As AI accelerators and high‑performance CPUs push past 1,000 W per chip and rack power exceeds 100 kW, traditional air‑based cooling is hitting its limits. A new white paper from IEEE Spectrum and Wiley, sponsored by CoolIT Systems, explains how single‑phase direct liquid cooling (DLC) can meet these thermal challenges. By routing water or water‑glycol through cold plates attached directly to…

1 source

Key points

  • Modern AI chips can exceed 1,000 W, and racks may dissipate over 100 kW, far beyond air‑cooling capacity.
  • Single‑phase direct liquid cooling circulates water‑based coolant through cold plates, removing heat faster and supporting higher density.
  • The white paper benchmarks DLC against two‑phase and immersion methods, positioning it as the most scalable solution for the next decade.

The paper compares DLC with two‑phase and immersion cooling, showing that the single‑phase approach offers a practical balance of cost, complexity, and scalability for the next ten years of ultra‑dense compute. It also projects future trends in processor power and rack density, underscoring the growing importance of advanced thermal design as AI workloads continue to intensify.

Full story from IEEE Spectrum AI · by CoolIT Open source ↗

Single-Phase Direct Liquid Cooling Is Proven for the Next Decade ofUltra-Dense Compute

IEEE Spectrum AI · 16 September 2026

Download this complimentary White Paper today!

This White Paper provides a comprehensive overview of how single-phase direct liquid cooling manages the rising thermal demands of AI and high-performance computing, and how it compares with two-phase and immersion approaches.

What you will learn about:

  • Why rising compute density has made heat the central design constraint in AI and high-performance computing, where individual processors now exceed 1,000 watts and racks dissipate more than 100 kilowatts.
  • How semiconductors respond to excess heat through thermal throttling, and why maintaining thermal margin supports higher performance and longer hardware life.
  • Why air cooling reaches its practical limit at high rack densities, and how liquid absorbs and carries away far more heat in a closed loop.
  • How single-phase direct liquid cooling works at the chip and system levels, and how it compares with two-phase and immersion cooling.
  • How processor power and rack density are expected to grow, and what these trends mean for the future of thermal design.

Click ‘LOOK INSIDE’ to download the PDF now.

LOOK INSIDE

Presented by IEEE Spectrum and Wiley, sponsored by CoolIT Systems

More Information

As computing systems move toward denser processors, tightly coupled server nodes, and higher-power racks, managing the heat they generate has become a defining challenge in data center design. Modern AI accelerators can dissipate well over 1,000 watts, and a single rack may release more than 100 kilowatts of heat. This is far beyond what air cooling can practically remove. Single-phase direct liquid cooling addresses this by circulating water or a water-glycol coolant through coldplates mounted directly on high-heat components. The coolant absorbs the heat and carries it away in a closed loop to a coolant distribution unit. Because liquid stores far more heat than air and removes it much faster, this approach supports higher chip and rack densities within a smaller footprint. This paper explains how single-phase direct liquid cooling works, how it compares with two-phase and immersion cooling, and how rising processor power and rack density are shaping the future of thermal

This text was published by IEEE Spectrum AI and written by CoolIT. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗

Topics · follow one to build your own front page
CoolIT SystemsIEEE SpectrumWiley

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us.

Comments

via GitHub Discussions

More in Hardware & Compute

All →

Related stories