Does India’s Data Centre Boom Have a Measurement Problem?

Over the past year, India has seen a surge in data centres. According to KPMG, India’s data centre capacity has expanded from around 375 MW in 2020 to nearly 1.5 GW in 2025, with projections to reach 2 GW by the end of this year. 

But does the electric power capacity—megawatts or gigawatts—accurately project a data centre’s strength? 

Equinix, one of the world’s largest colocation and interconnection data centre providers with 280 data centres across 36 countries, believes we may be looking at the wrong figures. It recently inaugurated its fourth data centre in Mumbai, MB3, without specifying the electric power capacity. 

Instead, the company emphasised that MB3, built with an initial investment of $95 million, delivers capacity for more than 1,370 cabinets. 

A cabinet, often used interchangeably with ‘rack’, is the standardised physical enclosure used to house servers, storage systems, and networking equipment. They are more or less akin to measuring the physical footprint of data centres, while also helping calculate the total weight loads and required rack space.

Cabinets Over Power

Across India, Equinix has invested $365 million to date, including a presence in Chennai. Combined, its five data centres offer a total capacity of 4,625 cabinets.

“We want to take the attention away from ‘megawatts’ to what you can actually do inside the data centre,” Manoj Paul, Managing Director of Equinix India, tells AIM.

He explains that cabinet-based measurement better reflects how different customers actually use a data centre, since power requirements vary widely across use cases.

While power is indeed a precise measure of consumption, they capture capacity rather than how that capacity is used or where value is created. For instance, a customer drawing as little as 3 kilowatts can still play a significant role in the digital economy. 

Paul points to telecom providers that may require minimal power for their deployments but deliver critical connectivity and services. For such customers, headline megawatt capacity is largely irrelevant, since their priority is not bulk power availability but reliable connectivity, low latency, and the ability to interconnect with networks and platforms within the facility.

Similarly, enterprises such as stockbroking firms may operate a mix of cabinets, some with higher capacity and many others running on just 3–5 kilowatts, yet collectively supporting substantial business activity.

Power is an incomplete metric for diverse, interconnection-driven workloads, and this matters in India. According to Mordor Intelligence, 84% of data centre capacity in India comes from colocation providers—third parties that lease out physical data space as well as ancillary services like power, cooling, and network connectivity. Here, power is a poor proxy for value.

Interconnection Over Compute

Equinix primarily operates as a colocation and interconnection provider rather than a hyperscaler. 

Instead of building and running its own large-scale cloud platforms like AWS, Google Cloud, or Microsoft Azure, it develops and manages data centres where enterprises, telecom operators, and even Big Tech companies deploy their infrastructure.

Equinix’s facilities are designed to enable dense interconnections between customers, making them hubs for network exchange rather than standalone compute-heavy campuses. Customers can directly connect to multiple networks, cloud providers, and partners within the same facility, reducing latency and avoiding reliance on intermediaries.

Paul explains that even hyperscalers operate two distinct layers—large server farms where compute and storage reside, and network nodes that sit closer to users and connect to telecom providers and enterprises. Equinix primarily caters to the latter, hosting these network nodes in locations optimised for proximity to metros, dense carrier presence, and low latency.

This distinction becomes even more relevant with AI workloads, particularly inference, where queries need to travel quickly between users and models. Locating these network nodes within highly interconnected facilities reduces latency, avoids unnecessary routing through intermediaries, and lowers bandwidth costs by enabling direct connections with a wide range of telecom providers and ISPs.

Paul explains that some fintech companies operate through small, distributed deployments across regions such as India, Dubai, New York, and Silicon Valley, using limited capacity in each location to deliver services globally. Their value lies not in how much power they consume in any one location, but how these distributed nodes work together to serve users globally.

“If you rate them by megawatts or kilowatts, they appear as very small loads, but their impact, both in the solutions they deliver and the revenue they generate, is highly disproportionate to the power they consume,” he says.

On the other hand, a single rack can be partitioned to accommodate multiple customers, sometimes as many as eight, with each running relatively small workloads but requiring extensive interconnection. In such setups, the number of connections can matter more than the amount of power consumed.

The Equinix executive asserts that evaluating customers solely on the basis of power capacity introduces a structural bias, in which those with higher kilowatt requirements are seen as more important, regardless of what they actually do.

“That’s why we focus on cabinets, interconnections, and how many connections a customer needs,” says Paul, adding that value is driven by how deeply customers are interconnected within the ecosystem.

Having said that, he does not dismiss power as a metric entirely, noting that it can still be a reasonable way to evaluate hyperscalers, whose operations are inherently power-intensive and more uniform.

However, he argues that applying the same lens to enterprise customers, which run varied, application-specific workloads in shared, interconnected environments, is fundamentally flawed compared with hyperscalers’ uniform compute clusters. This is because customers are willing to pay a premium for interconnection and ecosystem access, thus rendering power as the wrong choice for advertising. 

How Equinix Avoids Customer Concentration

Many large-scale data centre companies face significant risks due to customer concentration, as their revenue comes from select hyperscalers such as AWS. However, losing them due to unforeseen vulnerabilities—as seen with the strikes on AWS and Oracle data centres during the Iran conflict—translates into immediate revenue disruption.

For instance, in its draft red herring prospectus filed with the market regulator, Sify Infinit Spaces, which is India’s first pure-play data center provider, identified customer concentration as a key risk to its business.

The company relies heavily on a small group of buyers, with its top 10 clients accounting for 82.43% of its operating revenue in FY25. The top three clients alone accounted for 68.34% of that total. Any adverse shift in a major client's operating strategy, such as choosing to secure services from a competitor, could severely impact the company’s overall revenues.

Equinix, on the other hand, has ensured that its top ten clients account for just 16.1% of revenue. Paul says this balance is central to how customers evaluate Equinix, as diversification strengthens its financial stability.

“Our revenue is growing strongly, we’re highly profitable with an EBITDA of ~50%,” he notes. “Along with this, the fact that no single customer contributes more than 2.5% of our revenue gives new customers confidence that the business is not dependent on any one client.”

While hyperscalers do use interconnection, it is limited to specific parts of their infrastructure, such as internal compute and storage. As a result, allocating an entire data centre to a single hyperscaler can result in lower overall interconnection activity than a facility serving a large number of enterprise customers.

Paul asserts that Equinix has deliberately shaped its focus on enterprise and interconnection-heavy deployments rather than pure hyperscale capacity.