Synopsis
Data centres are significantly upgrading fire safety, cooling, and electrical systems following a recent Google Cloud incident. The surge in AI computing power, with AI racks costing millions, necessitates enhanced protection for these critical digital assets. Operators are investing in advanced fire suppression, robust power, and efficient cooling to prevent costly disruptions.Listen to this article in summarized format
Data centres, considered the backbone of a digital economy, have emerged as critical national infrastructure as the rapid use of AI triggers a sharp rise in computing power and value concentrated inside such centres. While a traditional server rack typically contains $50,000-$100,000 worth of equipment, an AI rack contains hardware worth $2 million-$7.5 million, per industry estimates.
A single AI data hall can house $500 million to more than $1 billion in computing infrastructure. An outage or cooling failure can potentially disrupt weeks of computing work, resulting in wasted compute, power, and time. Growing concentration of computing power and value is forcing operators to rethink ways to protect some of the country’s most critical digital assets, said experts.
Newer facilities are already making additional investments in fire suppression systems, upgraded electrical infrastructure, and higher fire ratings to handle the demands of AI workloads, said Sharad Agarwal, chief executive officer at Sify Infinit Spaces, a data centre operator.
Operators are also redesigning power and cooling systems to manage higher heat loads and accommodate lithium-ion batteries increasingly being deployed within racks.
“Even data halls now require higher fire ratings because many racks come fitted with lithium-ion batteries, a standard that previously applied only to battery rooms,” said Agarwal.
Beyond fire protection, operators are investing heavily in cooling and power infrastructure. “AI-ready facilities are increasingly deploying direct-to-chip liquid cooling, rear-door heat exchangers, and localised chilled-water loops to dissipate heat generated by dense graphic processing units (GPU) clusters,” said Sunil Gupta, CEO at Yotta Data Services. Operators are also building dedicated substations and custom power infrastructure to support large-scale AI deployments, he said.
Agarwal explained that summer heatwave conditions in many parts of India have long been factored into data-centre designs in India. However, the country’s climate and air-quality conditions continue to make cooling more challenging than the West, with average annual power usage effectiveness (PUE) at around 1.4 compared with about 1.1 in some parts of Europe and the US.
Data centre changes are also being driven by a sharp increase in computing density. Traditional enterprise racks typically operated in the 5-15 kilowatt range, while AI training racks commonly operate at 60-120 kilowatts, said Ashish Banerjee, senior principal analyst at Gartner. It indicates operators fitting significantly more computing capacity into the same physical space.
“Earlier, we could deploy around one megawatt of capacity in about 800 square metres,” said Agarwal. “Today, that same space can house around 1.5 MW, and in some cases close to twice that capacity.”
The value packed into these facilities has risen just as sharply. “The new-age GPUs are almost $40 million per megawatt,” he said, comparing it with $15 million-$20 million per MW for traditional cloud infrastructure.
Gupta noted that an individual AI rack today houses between $4 million and $7.5 million in hardware value. “A single high-density AI data hall routinely contains $500 million to over $1 billion in compute infrastructure alone,” he said.
The growing concentration of value is also changing how operators view outages and failures. “Disruptions no longer just affect services, they can delay model development, product launches and customer commitments,” said Banerjee. “Operators are spending more on power, connectivity, and cooling as AI shifts priorities from physical space to managing heat and electricity.”
With AI scaling from training to large-scale deployment over the next one to two years, the cost of disruptions will rise sharply as more business and consumer services start relying on AI infrastructure, experts said.
