TodayToday's edition
India's National Stock Exchange says its on-premises data centre now uses direct-to-chip liquid cooling, with 80+ GPUs this quarter and room for 224. Vendor, GPU model and kW per rack are undisclosed.
Energy & Cooling DeskEdited by Ege Erten
2 min readAI-assisted
NSE says its on-premises data centre will pass 80 GPUs this quarter, per NSE's statement as reported.
Illustration (AI-generated)
India's National Stock Exchange (NSE) said on Nov. 9 that it has deployed direct-to-chip liquid cooling in its on-premises data centre, with GPU capacity due to pass 80 this quarter and room for up to 224 GPUs across high-density racks.
NSE described the deployment as the first of its kind in Indian capital markets, according to Business Standard, Business Today and ANI, which all carried the exchange's statement. The exchange said it already runs multiple GPUs and that the cooling architecture is built to scale. It also plans to extend the technology to its co-location data centre, where market participants host their own servers, in the near future and depending on their requirements.
The 80-plus GPUs planned for this quarter are just over a third of the 224-GPU ceiling, so most of the liquid-cooled capacity NSE has designed for is still empty. The 224 figure is a design ceiling, not a purchase order, and NSE did not say when it expects to reach it.
NSE said the on-premises setup is meant to strengthen data sovereignty and let it keep complete control over AI data, computing resources and access, in line with its security governance and applicable regulatory requirements. Viral Mody, NSE's Executive Director, said the exchange is investing in technologies that improve efficiency and resource use. Chief Technology Officer Sampath Manickam said AI is driving a fundamental shift in infrastructure, with significantly higher demands on compute density, power and cooling, and called the deployment a scalable and energy-efficient foundation. Those efficiency statements are NSE's own; no power, PUE or water figure was published.
At 224 GPUs the site is small beside hyperscale halls, and the reports do not give the rack count, so the density per rack cannot be worked out from them. For a sense of what direct-to-chip loops are now expected to handle, Voltgrid reported a Korean lab that reproduced 140 kW racks. The NSE deployment matters less for its size than for who is buying: a regulated market operator is putting liquid cooling inside its own walls rather than renting capacity.
NSE has not named the cooling vendor, the GPU model, the rack density in kW, the facility water temperature, the site or the cost, and none of the reports state them. It has not given a date for reaching 224 GPUs or for the co-location rollout. The efficiency claims have no published basis. Watch for a vendor or integrator announcement and for any PUE figure once the first racks run under load.
Cooling & rack power
Cooling & rack power
Cooling & rack power
Cooling & rack power
