LiquidStack’ s GigaModular CDU delivering direct-to-chip liquid cooling with up to 10MW capacity
The situation is exacerbated by timing mismatches between AI development cycles and traditional data centre construction.
“ Typical design build cycle is at least 18 months,” he points out.
“ Based on the initial tailwinds around Blackwell, the new AI purpose-built data centres are only really coming online in Q3.”
This disconnect has forced many operators to retrofit existing air-cooled facilities – a process Joe describes as“ very disruptive” since these facilities weren’ t designed for the necessary piping infrastructure.
The challenge is compounded further by operational temperature differences: whilst liquid cooling systems optimise for 40-degree Celsius water, existing facilities typically operate at 20 degrees, leading to higher power usage effectiveness ratios.
The competitive advantage of specialisation In an industry where most major players offer cooling as one component of broader portfolios, Liquid Stack’ s singular focus provides distinct advantages.“ This is all we do,” Joe says. aimagazine. com 31