DELSKA
CREDIT: XXX
from 50 to 250 kilowatts. This makes capacity planning extremely difficult.
Chief Technology Officer Rihards Kaletovs designed the new facility in Latvia to handle both scenarios without wasting money on unused capacity.
“ Built on a modular design, the site scales from 10 MW to 30 MW on owned land next to the construction site, enabling right-sized, on-demand capacity without overprovisioning,” he says.
The facility uses two cooling systems. Standard air-cooling handles regular servers, whilst liquid cooling manages high-power AI chips that would otherwise overheat.
“ Our dual-mode cooling setup supports both traditional air-cooled systems and high-density liquid cooling loops, designed explicitly for GPU-intensive AI and ML clusters,” says Rihards.
Andris explains why this flexibility matters for business planning.
“ At the simple level of regular hosting, it’ s 10-20 kilowatts per rack,” he says.“ When it comes to AI, it’ s a complete disaster in terms of predictability. You can have 50 kilowatts, 100, 250.”
Building separate facilities for different power requirements would cost too much. Instead, Delska designed a new facility that handles both.
“ You can’ t build a data centre capable of hosting a 10-kilowatt rack and a one-megawatt rack,” says Andris.“ So, when you are planning server rooms or a data centre, you need to plan these levels and ask, what is the upper limit for you?”
Connectivity features at Delska’ s new facility include direct cloud connections, low-latency links and a carrier-neutral meet-me room where multiple network providers can connect.
“ The facility empowers customers to build hybrid, high-performance environments with zero vendor lock-in,” says Rihards.
The facility is set to open in autumn 2025, with land available for expansion and power capacity already secured for up to 30 MW for future deployments.
sustainabilitymag. com 125