Canadian liquid cooling vendor CoolIT Systems has demonstrated what it claims is the first 15 KW cold plate design that delivers four times the performance of earlier single-phase direct liquid cooling (DLC) platforms.
The company made the announcement during Computex 2026 taking place in Taipei, Taiwan, this week.
The company said the demonstration shows that single-phase DLC can scale to meet the thermal demands of future GPUs and AI accelerators. Something that has been called into question as these technologies begin to accelerate in importance in AI data centers.
CoolIT is pushing back on the idea that cold plate design may not be able to handle the thermal demands of these future chips. According to a report from IDTechEx, with thermal design requirements continuing to rise, the industry is reaching the limits of what single-phase cooling can handle. Two-phase direct-to-chip (D2C) cooling is expected to enter large-scale deployment in 2027.
The report also said that while the exact timeline may vary, a consensus is that single-phase D2C begins to struggle at about 1.5 kW, with 2.0 kW seen as the upper limit.
CoolIT’s announcement suggests that cold plate technology still has lots of life left and could still meet AI GPU and accelerator demands through 2030, moving beyond its purported upper limit to 15 kW.
Nvidia’s Vera Rubin AI data center platform, now in full production, has a thermal design power (TDP) of up to 2.3 kW per GPU, already exceeding the 2.0 kW threshold identified as the practical upper limit for single-phase cooling. CoolIT’s responding to a chip that is already in production and already pushing past where single-phase was supposed to tap out.
“Single-phase DLC is already cooling millions of AI accelerators today. This achievement shows it is also the architecture to cool AI infrastructure well into the future,” said Kamal Mostafavi, CTO of CoolIT Systems. “With validated performance at 15 kW, CoolIT has proven that single-phase DLC is not only practical to cool millions of the most advanced AI chips today — but ready to cool the coming generations of GPUs and AI accelerators.”
Why it matters
On-chip cooling will at some point compete with cold plates but not in the near future.
Cold plates are mounted on GPUs, CPUs and sometimes memory modules where coolant flows, absorbing the heat and circulating through rack manifolds and coolant distribution units in a closed loop. This allows for more precise thermal control than room- or rack-level cooling. Cold plate design shows more than 97% heat removal for high-power chips.
These 15 kW cold plates will allow the company to offer 10 times the performance of current models and will give the company a pathway to continue to support cold plate technologies to at least 2030.
CoolIt also said that Nvidia has highlighted single-phase DLC with 45° C supply temperature as part of its next-generation AI platform roadmap. This confirms that warm-water liquid cooling and cold plate technologies will continue to be needed in the future for factory-integrated systems.
According to a blog post from Schneider Electric, single-phase cold plate cooling will be the default choice and likely to dominate in the next five to 10 years due to its ease of deployment, lower cost and alignment with future roadmaps. But after 2030, two-phase liquid cooling will begin to gain traction.
Overall, the global data center liquid cooling market is projected to rise from $5.7 billion in 2026 to $29.2 billion by 2033, seeing a compound annual growth rate (CAGR) of 26.4% during the forecast period, according to data from Persistence Market Research.
