3.5 GW of Data Centres Got Announced. The Substations Did Not.

India's data centre industry is forecast to grow 66% by 2026, with capacity projected to expand another 30% year-on-year and around 500 MW of fresh supply expected online this year alone. Between March 2025 and April 2026, operators announced roughly 30 large projects collectively adding 3.5 GW of planned capacity, with Andhra Pradesh and Telangana alone accounting for over 2 GW of AI-centric mega campuses. What gets far less coverage than the announcements: a hyperscale AI data centre's power demand is an entirely different order of magnitude from a traditional enterprise facility, and grid infrastructure — substations, transmission capacity, HT connections — moves on a timeline measured in years, not the quarters an announcement press release implies.

Why AI data centres break the old power-planning assumptions

• AI/GPU compute density draws dramatically more power per rack than traditional enterprise server infrastructure — a data centre sized for "servers" in the old sense can be an order of magnitude under-provisioned for an AI-workload campus of the same floor area
• Announced capacity figures describe the campus plan, not the grid connection already secured — a 40MW facility announcement doesn't mean a 40MW substation connection exists or is even approved yet
• Redundancy requirements multiply the real demand — the N+1 or 2N redundancy a serious data centre needs applies to power infrastructure just as much as cooling, meaning the grid connection required is meaningfully larger than the facility's nameplate IT load alone

Where the power bottleneck actually bites, project by project

Why the tax incentives don't solve the physical bottleneck

The Union Budget 2026-27's 20-year tax holiday for eligible foreign cloud providers and 25-35% capital incentives for green technology adoption make the financial case for building in India stronger than ever — but tax policy doesn't build a substation. The capital incentives and the physical grid infrastructure timeline are two entirely separate constraints, and an operator that's solved the financing case still has to solve the power-delivery case, often on a longer critical path than the building itself.

What a realistic data centre power strategy actually requires

1. Grid capacity confirmed with the actual utility before site commitment, not assumed from a general regional power-surplus narrative — the specific substation and feeder serving that exact site is what matters
2. HT connection and substation upgrade applications filed in parallel with, not after, construction planning — given multi-year utility timelines, this is very often the true critical path for a large campus, not the building itself
3. Full-campus power demand modelled from day one, even for phased construction — the same full-build-out logic that applies to shared campus infrastructure generally, so Phase 2 and 3 aren't stranded behind an undersized initial grid connection
4. On-site generation and storage evaluated as a genuine part of the power strategy, not just a backup — captive generation, solar and battery storage can bridge gaps while grid infrastructure catches up, particularly relevant given the DG-vs-grid-vs-solar economics at this scale
5. Redundant power paths designed against the facility's actual uptime tier requirement, not a generic assumption, mirroring the same discipline needed for cooling redundancy

Our electrical infrastructure team works through grid-capacity verification and HT connection strategy at the site-selection stage — before capital is committed to a location the grid can't actually support at the announced scale.

FAQs

Why does a data centre's announced capacity not match its actual available power?

Announced capacity typically describes the campus's planned build-out, not a confirmed grid connection — substation and transmission capacity at that specific site needs separate verification with the utility, and often takes longer to secure than construction itself.

How much more power does an AI data centre need compared to a traditional one?

Meaningfully more per rack, given GPU/AI compute density — a facility sized using traditional enterprise server assumptions can be significantly under-provisioned for an AI-workload campus of the same floor area.

Can on-site generation solve a data centre's grid capacity gap?

It can bridge a gap while grid infrastructure catches up, but captive generation, solar and storage need to be sized and integrated as a genuine part of the power strategy, not treated purely as backup for an otherwise-adequate grid connection.

Can you assess grid capacity and power strategy for a data centre site before we commit to it?

Yes — site-specific grid verification and HT connection strategy as part of site selection. Request an assessment.