12 hours ago
Data centres in non-metro cities a non-starter as AI warrants scale over speed
New Delhi/Mumbai: Edge data centres, which were expected three years ago to boost India's cloud services market with proliferation in smaller cities, have remained a pipe dream as applications requiring high connection speeds have failed to take off as predicted.
These smaller data centres, designated to be less than 10 megawatt (MW) in capacity as against a hyperscaler facility's 50MW-plus size, serve local data demand within districts. They provide high connectivity speeds, or low latency in technical parlance, by reducing the physical distance to the end user. They were planned to come up in tier 2 cities such as Guwahati, Patna, Lucknow, Jaipur, Nagpur, Pune and others to serve applications like smart cars, augmented reality headsets, automated traffic management and more.
However, these applications have not scaled as predicted by analysts, resulting in edge data centres falling out of favour after some initial growth spurts. After a brief burst of enthusiasm, operators are pouring money into expanding large facilities in established hubs like Mumbai and Chennai, where enterprise and big-tech demand is deep and revenue visibility stronger.
Icra Ltd estimates India's edge capacity will rise from 70MW in 2024 to about 200MW by end-2027—just 8% of the projected 2.5GW total. Anupama Reddy, vice president and co-group head of corporate ratings at Icra, noted the barriers for edge: security risks from remote deployments, rapid tech shifts and obsolescence risk, talent scarcity, and interoperability challenges with core data centres. She added that rentals are structurally higher for edge, reflecting higher capex per unit of data and a retail-heavy customer mix.
Real estate consultant JLL, in a report in June, concurred with Icra's analysis.
Hyperscalers up speed
Meanwhile, hyperscale—or large–data centres—have been cutting down on latency, eating into the key selling point for edge data centres, said Sanchit Vir Gogia, chief executive of technology advisory firm Greyhound Research. 'The industry miscalculated how quickly core network improvements and submarine cable projects would help hyperscale data centres close much of the latency gap," he said.
As of today, only a fifth of chief information officers at enterprises report active investment in edge data centres beyond primary metros, as per Greyhound. 'After years of hype, deployments have lagged due to the absence of widespread latency-critical workloads, and the rapid maturity of backbone networks and data distribution architectures," Gogia said.
Reddy and Gogia's projections are backed by the data centre firms, too. Operators such as Singapore-headquartered Princeton Data Group, Hiranandani-backed Yotta, Bharti Airtel's Nxtra and US-based Equinix, among others, are all looking to build large data centres—instead of edge facilities.
'Edge data centres are not a part of our strategy, and we expect hyperscalers to be the largest demand driver," said Vipin Shirsat, managing director for India at Princeton Data Group. Use cases of edge data centres include content distribution and low-latency applications like gaming or autonomous cars, he said. While these use cases are growing, they don't really require significant capacity, he explained.
'The uptake of these applications is not what many had anticipated three years ago," Shirsat said.
'Applications that require lower latency are rare, and tier 2 locations face hurdles such as higher per-MW build costs for edge facilities, limited redundant networks, and a smaller pool of skilled technical talent," said Manoj Paul, managing director of Equinix India, adding that the company will continue to concentrate its investments in Mumbai and Chennai.
Edge lacks demand
Sunil Gupta, chief executive of Yotta, echoed Shirsat and Paul—stating that while the company 'already has the land in various cities for edge expansion, the client demand isn't at a scale that warrants us to invest in edge data centres right away."
'When demand builds, we'll definitely ramp up based on client needs," he added.
Bharti Airtel-owned Nxtra, meanwhile, is different from the rest. Ashish Arora, chief executive of Nxtra, said that the company currently has 150 edge data centres in 65 locations, with plans to add 10 more in the next two years. 'For edge facilities, most of the demand comes from within the Bharti group, but in terms of our overall business, clients from outside of the group now contribute to the majority of our revenue," he said.
Even then, Arora said that edge data centres are not a mainstay of the company's data centre growth plan. 'Edge data centres won't be the industry's primary growth driver going forward—tech-driven services are still seeing investments made towards innovation, and consumer-end demand is limited. Hence, for data centres, most of the demand will be for traditional data centres in established hubs," he added.
Arora further said that AI applications have not created demand in smaller districts, for now. 'You may want lower latencies in consumer-end AI applications, and in inferencing from running AI models. But, training AI models—the main data centre use case—needs large capacities and not fast deliveries, which is why such use cases are very low and limited for now," he added.
Companies that set up large data centres have announced expansion plans such as Princeton Data which, in September last year, said it would establish 230MW in net India data centre capacity, with the first phase targeted to be brought live by the end of next year. Equinix, too, announced a large-scale expansion plan last year. Yotta's Greater Noida facility, announced in October 2022, can expand up to 250MW by next year. However, there are no specific expansion plans for edge data centres.
As a result, industry stakeholders believe there are little to no business benefits for edge data centres for now. As Greyhound's Gogia said, 'A national edge rollout plan was shelved after content distribution upgrades reduced data latency significantly. While edge data centres delivered initial gains, their incremental advantage could not justify the additional capital and operational burden."
