Share with your CTO
Brookfield is placing a direct bet on AI inference infrastructure in India, projecting the country will add 6 GW of data centre capacity over the next five years, up from a current 1.5 GW base that Arpit Agrawal, Brookfield’s head of India infrastructure, describes as “almost all non-AI.” The firm is pairing that infrastructure play with renewable energy, while explicitly building in risk mitigation for curtailment, the problem where weak or congested grid infrastructure forces clean power plants to shut off output they could otherwise deliver.
What this means for your business
The 6 GW projection is the number worth stress-testing. India’s entire current installed data centre capacity sits at 1.5 GW. A 4x expansion in five years, concentrated in AI inference workloads, is an infrastructure commitment at a scale that changes vendor economics, real estate timelines, and talent geography for any enterprise running regional AI deployments across Asia-Pacific. If you’re evaluating where to anchor inference capacity outside the US and Western Europe, the question is no longer whether India is a viable option but whether your planning cycle is moving fast enough to secure early positioning.
The curtailment risk disclosure is the most analytically interesting part of what Brookfield is saying. Curtailment is normally a renewable energy finance problem, but at the scale of AI inference data centres, it becomes a latency and availability problem too. A facility powered by renewables in a region where grid operators regularly force plants offline is a facility with unpredictable power budgets. Brookfield is signaling that it’s pricing this risk in rather than papering over it, which is a more credible posture than the typical “we’re building sustainably” framing that elides how immature many emerging-market grids actually are for always-on compute loads.
The falsification condition for Brookfield’s thesis is grid development speed. If India’s transmission infrastructure doesn’t keep pace with data centre construction, the curtailment risk doesn’t stay manageable through clever contracting; it becomes a structural cap on utilization rates. CTOs evaluating India-based inference capacity should treat grid buildout milestones, not just data centre commissioning dates, as the leading indicator worth tracking in their vendor due diligence cycles.
Concept deep-dive: Curtailment
Curtailment happens when a power grid operator instructs a generating plant, typically wind or solar, to reduce or stop output even though the plant could produce more, because the local transmission network can’t absorb the electricity without risking instability. For data centres, which need continuous, predictable power, a co-located or grid-connected renewable source subject to frequent curtailment means the facility either draws on fossil backup or faces genuine uptime exposure. It’s the hidden cost in “powered by renewables” claims attached to infrastructure in grid-constrained markets.
Based on reporting from Brookfield bets on AI inference, clean power in India while tackling curtailment risks, originally published 2026-07-28 07:30:00.
