In a note to clients Tuesday, analysts at Mizuho provided their takeaways from last week’s Computex event. They noted that server liquid cooling was the main focus outside of the releases of updated AI GPU roadmaps from NVDA, AMD (NASDAQ:), and INTC. In addition, the investment firm highlighted the market forecasts from SMCI.
Firstly, Mizuho said most server SI (System Integration including ODMs and OEMs) and thermal solution vendors demonstrated their technologies in the exhibition.
They visited over 10 company booths regarding server cooling products at Computex, with a focus on DLC components such as CDU, manifold, and cold plates in server racks. DLC supply chain includes thermal solution providers (components), enclosure vendors, system integrators (ODM/OEM), and data center infrastructure vendors.
“Power efficiency in data center infrastructure becomes an urgent issue in the HPC/server industry and is likely to be a bottleneck for AI server growth in the near future,” said Mizuho.
They added: “We learned from an industry expert that if all data center servers in the world switch to accelerating computing architecture (with AI servers installed) from general purpose computing, power consumption will be close to 140x the current level, which looks unachievable in the next 5 to 10 years.”
Mizuho believes the development of DLC and liquid immersion solutions is at an early stage and that opportunities remain open to all supply chain vendors in the server industry. However, they note it is a highly competitive, fragmented market with “massive capex/R&D spending needed to take market share, as current leaders are more from the US, close to Nvidia/AMD.”
As a result, they expect more industry collaborations and consolidations across the region and the relationship between DLC components and SIs to be further strengthened.
For SMCI, Mizuho said the company’s CEO, Charles Liang, presented the outlook for DLC in the AI server industry.
“SMCI expects DLC penetration rate in new DC (data centers) to rise sharply from 1% in 2023-2024 to 15% in 2024–2025, and 30% two years from now,” stated Mizuho.
SMCI is said to believe that its DLC solution, compared to current air cooling for AI GPU servers, saves 0% to 3% of capital spending for one new DC equipped with 1,000 HGX systems and up to $60m million savings in OPEX over five years based on 33% lower CO2 emissions.
“The lead time of DLC server orders at SMCI has shortened to 2–4 weeks, based on the company’s current 1,000-unit monthly capacity,” says Mizuho.
