The government's push for sustainability in the IndiaAI Mission graphics processing units (GPU) tender, which requires bidders to have a power usage effectiveness (PUE) of less than 1.35, has data centres concerned because it will force them to increase capital expenditure and, as a result, pass the cost on to customers.

The government's procurement wants to establish public AI cloud infrastructure with 10,000 or more GPU capacity under a public-private partnership format.

The PUE statistic assesses how effectively a data center consumes energy. It is computed by dividing a data center's overall power consumption by that of its computer equipment. Lower PUE ratios indicate more energy efficiency.

According to a new condition established by the Ministry of Electronics and Information Technology (MeitY) in a corrigendum to the offer released in the final week of September, a bidder must affirm that the PUE for the data centers from which AI services on cloud will be supplied is not greater than 1.35.

Some industry leaders believe this condition is difficult to meet.

"The PUE requirement should be deleted. "There are multiple technologies, and each will have its own PUE," a data center executive told ET. "Regardless, GPU-as-a-service providers will try their utmost to eliminate PUE because it is a cost item for them. People would use technology with the quickest response time in the beginning, according to the source.

"Also, 1.3 PUE is not possible in India with direct-to-chip liquid cooling due to technical reasons like high ambient temperature," the company's boss stated. "In our view, PUE criteria should be avoided."

One GPU server requires a power density of 10 kW. According to experts, cooling a 10-kilowatt server traditionally takes another 10 kilowatts.

Water-based air conditioning units or dry coolers are used to chill data centers, which serve as the cornerstone for artificial intelligence infrastructure. Dry coolers need more power, but water-based coolers use more water.

This technique is not sustainable in a hot tropical nation like India, where certain areas lack access to clean water. Dry coolers, which take more power, are also not sustainable.

"There are data centers throughout India, and some of them are underutilized. By including this provision, you are not using them to their maximum potential," stated an official at a cloud computing business. "Old data centres will not be able to handle this PUE. This is similar to bringing Western norms on India," the source stated.

"Data center costs will rise, and we will pass those costs on to customers," the CEO stated.

According to Mordor Intelligence, the major firms in the India data center market include Equinix, NTT, Nxtra, Sify, and ST Telemedia Global Data Centres.

According to Narendra Sen, CEO of Indore-based NeevCloud, which has a patented liquid immersion cooling technology, cooling is one of the major cost factors for data centres, aside from GPU operation, and there are several types of cooling technologies available: liquid cooling, direct-to-chip liquid cooling, immersion cooling, and active rear door cooler (RDHX).

He claimed that the PUE of data centres in India is now about 1.5.

"With direct-to-chip liquid cooling, you can easily achieve 1.2 PUE," Sen told ET, speaking as an independent expert who is not participating in the contract. "Yes, we're in a tropical environment, but a 1.5 PUE will increase the customer's cost. Using water for cooling is not sustainable in a water-scarce nation. Traditional data centers must update."

He said that, in addition to space occupied, data centers use PUE as the second charging component for clients when putting up racks. "Singapore also has similar climatic conditions, but their data centres' PUE is less than 1.3," he noted.

In a YouTube video posted last year, Xavier Kurian, director of solutions and alliances at Dell Technologies, stated, "PUE in India is normally around 1.6, which implies that if you have 1.6 MW of power in a data center, 1 MW will be used for IT equipment. At a larger scale, this provides a chance to enhance PUE while saving power.

H200 GPUs are coming shortly to E2E Cloud

E2E Networks, an NSE-listed artificial intelligence cloud services company, stated on Tuesday that it will be the first in India to provide Nvidia H200 GPUs, which deliver 45% more performance than H100.

"Get ready to supercharge your model training with the latest and most powerful GPUs available," E2E's chief revenue officer, Kesava Reddy, said on LinkedIn. "At E2E Networks Ltd, we are dedicated to accelerating India's AI journey by offering access to cutting-edge technologies. Innovation flourishes when you have the greatest tools, and we're here to make sure you have them. Stay tuned for further developments. Let us grow together in the AI revolution.