Global data centers are running near full capacity, and the four biggest US hyperscalers are on pace to spend roughly $700 billion this year and still can’t build fast enough to keep up.
That capacity crunch is exactly what’s pulling capital toward India.
At February’s AI Impact Summit in New Delhi, Adani Group and Reliance Industries jointly pledged $210 billion in AI infrastructure investment, and PwC now projects India’s data center ecosystem could unlock a $280 billion opportunity by 2035 as capacity expands roughly 8.5 times over.
Where the Money Is Actually Going
Adani is committing $100 billion through 2035 to grow its AdaniConneX platform from about 2 gigawatts toward 5, funded in part by its 30-gigawatt Khavda renewable energy project, more than 10 gigawatts of which is already running, alongside a $15 billion joint venture with Google building a 1-gigawatt AI hub in Visakhapatnam, expected to be Google’s largest AI campus outside the US.
Reliance and Jio are putting in $110 billion over seven years, with chairman Mukesh Ambani describing three pillars: multi-gigawatt AI and digital infrastructure, an in-house renewable energy surplus anchored in Kutch and Andhra Pradesh, and a nationwide edge-compute layer built into Jio’s existing network.
Microsoft reaffirmed its $17.5 billion India commitment at the same summit, part of a broader $50 billion Global South investment plan, with its Hyderabad site going live mid-2026.
On the software side, the government’s IndiaAI Compute initiative is adding 20,000 subsidized GPUs, while Anthropic signed an enterprise partnership with Infosys for Claude deployments and OpenAI formalized its “OpenAI for India” program, layering model access on top of the physical buildout.
The Bottleneck That Could Slow the Rush
The binding constraint isn’t capital, it’s the grid. India is adding renewable energy faster than the grid can deliver it. About 300 gigawatt-hours of clean power was curtailed during the first quarter of 2026 because transmission infrastructure could not carry it.
Around 40 gigawatts of auctioned renewable capacity also remains without signed power purchase agreements. Financially stressed state distribution companies remain reluctant to commit to long-term contracts.
That matters more now than it used to, because AI infrastructure changes what a data center actually draws.
A conventional server rack typically consumes 5 to 10 kilowatts of power. AI training racks require 20 to 30 kilowatts, while the largest clusters can approach 100 kilowatts. Existing transmission infrastructure was not designed for this level of demand.
There’s a genuine upside hiding in that risk, though: hyperscalers and conglomerates like Adani and Reliance carry investment-grade balance sheets, which can lower borrowing costs for renewable projects across the sector once enough of them sign on, potentially making clean power cheaper for everyone, not just the data centers driving the demand.
The capital committed this year answers the easier question. Whether India’s grid, regulation, and transmission timelines can move fast enough to actually use it is the one still open.
Source: Official Moneycontrol AI Edge Newsletter, "India's AI Data Centre Gold Rush"
