“We have an opportunity to build arguably the biggest infrastructure build-out in the history of the planet,” said Santosh Janardhan, global head of infrastructure at Meta Platforms, who is leading CEO Mark Zuckerberg's ambitious ‘Meta Compute’ initiative to build tens of gigawatts of AI data centre capacity over the next decade.

The executive, who oversees Meta’s $115-$135 billion annual spending on AI compute, told ET’s Himanshi Lohchab that the company has huge infrastructure plans for India. “I was in Jamnagar (Gujarat) yesterday with Aakash (Ambani) and the work there is underway at record speed,” he said referring to the 168 megawatts data centre deal with Reliance Industries.

Janardhan also highlighted the growing importance of Meta's Bengaluru engineering hub, revealing that teams there play a key role in developing the firm's proprietary silicon, undertaking highly specialised work that “the rest of the world frankly can't do.” Edited excerpts:


What is Meta’s infra plan for India?

Through our family of apps, we touch 3.5-4 billion people regularly. We are in a transformational era. We have the distribution, and when you look at what AI can enable on top of that distribution, the possibilities are extraordinary. India is a huge part of that story. We recently announced the data centre project in Jamnagar. I was there with Akash and the work is underway at record speed. But if you step back, there is a broader India narrative.

We have Project Waterworth, which is the world's longest and most sophisticated subsea cable system. India is a huge part of that. We also have a significant presence in Bengaluru where we build custom silicon. A meaningful portion of Meta's revenue comes from silicon that we design ourselves. A big part of that work happens in Bengaluru. We have engineers there doing highly specialised work that, frankly, the rest of the world cannot do. Our ambitions in India over the next 5-10 years are enormous.

What investments has Meta made in India over time?

About 5-10 years ago, our focus was on enabling traffic in India. India was one of many growing markets and we wanted to serve users locally rather than backhauling all traffic back to the US. What is unique about India is that Meta's growth and the country's digital growth happened simultaneously. The demographics shifted, internet adoption increased and demand kept rising. Almost all the graphs are going up and to the right. That's a rare phenomenon. We started with edge infrastructure and points of presence (PoPs). Then we moved to mega PoPs because the scale of content consumption in India required a different architecture. India is one of the few countries where we have mega PoPs in the sense that edge networks alone don't cut it. We needed to cache local content differently than the rest of the world.

What are the scaling plans for Jamnagar data centre?

It's still early days. This is our first foray. In AI, there is a term called "pipe cleaning." It means doing your first zero-to-one project, learning what works, getting all the ingredients right and then scaling. That's how we're approaching in this project. There are a lot of reasons to go very big in India, but you walk before you run.

Will the DC be used for AI training or inference?

I don't like to think of data centres as being built for one specific purpose. Can we train there? Yes. Can we do inference there? Yes. Can we serve the family of apps there? Yes. All of those are possibilities. I would encourage people to think about data centres as the factories of today. Today's data centres take in data and compute. The output is intelligence.