There’s a hard number sitting behind every AI roadmap, and it’s measured in electricity. Global data-centre power use is on track to roughly double, from about 415 TWh in 2024 to around 945 TWh by 2030, close to 3% of all the electricity used on Earth (IEA). AI is the main reason. For an industry built to keep growing, that raises an uncomfortable question: can you scale compute without blowing past your climate targets?
At Vyoma.ai, we don’t see that as a trade-off, and we don’t treat green as a finishing layer added once the concrete is poured. In India especially, sustainability isn’t the polish at the end. It is the thing that decides whether you get to build at all, because the grid, the water and the land here won’t let you brute-force your way to scale. So we design our data centres to be efficient first, engineered in from the power feed to the cooling loop, because at hyperscale the things that make a facility greener are usually the same things that make it cheaper to run and faster to get powered.
The Sustainability Bill Coming Due
The efficiency gap is wider than most dashboards admit. Industry-average PUE has barely moved in years; the Uptime Institute puts it at about 1.56, essentially flat for half a decade (Uptime Institute). A 1.56 means that for every watt doing real compute, you spend more than half a watt again on cooling, power conversion and overhead. The best hyperscale sites run near 1.1 (Uptime Institute) — on those figures, an average facility burns more than five times the overhead energy of a best-in-class one. Most of the world’s installed base is nowhere close.
Water is the quieter problem. Only 41% of operators even track it (Uptime Institute), and US data centres alone drew an estimated 17 billion gallons in 2023 for cooling, a figure that could double by 2028 (LBNL). Carbon is the third, and it follows whatever grid the site happens to sit on.
India is about to meet all three at once, and harder than most. Our data-centre power demand is projected to grow nearly fivefold, to 57 TWh by 2030, lifting the sector’s share of national electricity from 0.8% to about 2.6% in six years (S&P Global). Installed capacity has to race from roughly 1.5 GW today toward 4 to 5 GW by 2030 (IBEF). But the grid can’t just be asked for more on demand: in parts of Maharashtra, a fresh connection above 10 MW already carries an 18-to-24-month wait (Mordor Intelligence), and the metros driving the demand are the same ones where water is tightest. Build all of that the old way and the grid, the water table and the carbon math each take the hit. This is the real reason efficiency matters here. Every watt lost to overhead is a watt that never reaches a GPU, and on a constrained grid that isn’t just a worse carbon number, it’s compute you will never get to build.
Why We Build Green by Design
Retrofitting efficiency into a finished building is slow and costly. Designing for it from the first drawing is neither. And there is hard evidence the bolted-on approach fails: industry PUE has been stuck at about 1.56 for five straight years (Uptime Institute), even as nearly every operator published a sustainability report over the same stretch. A certificate bought after commissioning doesn’t move that number. What moves it are decisions taken before the first line is drawn — how power enters the building, how heat leaves it, how the whole hall is instrumented.
That is the part an engineering-and-construction company is built to get right. L&T doesn’t buy these campuses finished and badge them green; we design and build them, so efficiency goes into the foundation drawings, the substation and the cooling topology, the places you cannot retrofit later without tearing the building apart. It also matches where the country is heading. Green-certified data centres are about a quarter of the Indian market today and are expected to reach 30 to 40% by 2030 (IBEF/Colliers), while states like Maharashtra, Tamil Nadu and Karnataka now tie data-centre incentives to using at least 30% renewable energy. The regulatory floor is rising, and we’d rather sit well above it than scramble to meet it.
Energy: GIS Power, Low PUE and Real-Time Control
Efficiency starts at the fence line, with how power gets into the building. Our Chennai campus runs on an in-house GIS (gas-insulated switchgear) substation. GIS is compact and loses less energy than a conventional open-air switchyard, so it trims both electrical losses and land footprint before a single server is switched on. Inside, integrated BMS and DCIM systems watch power and cooling continuously and tune them as load shifts, which is how a facility actually holds a low PUE in daily operation rather than just on a design sheet. We engineer for best-in-class PUE, well under the 1.56 industry norm, and we’d rather prove it in running facilities than quote a single number on a brochure. The payoff isn’t only a lower carbon figure: because a lower PUE puts more of every megawatt onto the GPUs, on India’s constrained grid that efficiency converts directly into capacity we can actually sell — headroom a wasteful facility never gets to build.
Cooling and Water: Liquid Where It Counts
Cooling is where a lot of the waste hides. In older facilities it can swallow 30 to 40% of total energy (IEA); in efficient hyperscale designs it’s a small fraction of that. As AI pushes racks past 100 kW, air simply can’t keep up, so the high-density halls behind our AI Factory are built around liquid. We bring Direct Liquid Cooling to the chip on the densest systems, run Rear Door Heat Exchangers on mixed-density floors, and move the most thermally aggressive workloads into Liquid Immersion Cooling. Shifting from all-air to largely liquid cooling can cut facility power by around 15% on its own (LBNL) — efficiency and thermal headroom in the same move.
Liquid also changes the water equation, and in India that matters more than any global average lets on. Evaporative cooling drinks water; US data centres alone consumed an estimated 17 billion gallons in 2023 (LBNL). In water-stressed Indian metros, that draw isn’t just an ESG line — it’s a siting risk and a question of social licence, because a data centre and a city can end up pulling from the same stressed source. Our cooling runs on closed loops that recirculate fluid instead of evaporating it away, keeping Water Usage Effectiveness (WUE) low. Here, closed-loop isn’t a nicety; it’s often the difference between a site you can build and one you can’t.
Carbon: Certification, Scope 2 and Clean Power
The last front is carbon, and most of it comes down to the power source. Our Chennai campus is being built toward LEED certification, which holds the design to independent benchmarks across energy, water and materials rather than a self-declared claim. For Scope 2, the emissions tied to the electricity we buy, the answer is clean-power procurement, and India is a good place to do it. The country is targeting roughly 500 GW of non-fossil capacity by 2030 and lifting renewables to about 32% of generation from 13% in 2024 (S&P Global). Round-the-clock renewable PPAs can also cut operating costs by as much as 30% against grid tariffs (Mordor Intelligence), so the cleaner supply is frequently the cheaper one too. That’s the pattern running through this whole piece: here, the green choice and the commercial choice keep turning out to be the same choice, which is why we treat clean procurement as margin, not philanthropy. And because L&T builds these campuses itself, we can design embodied carbon down at the structural stage, where concrete and steel can account for a large share of a data centre’s lifecycle emissions (World Bank) and where a pure colocation operator has no lever at all.
Conventional vs. Green-by-Design
Set against a legacy build, the gap isn’t a plaque or a tariff. It is where efficiency gets decided.
|
Dimension |
Conventional Data Center |
Vyoma.ai Green-by-Design |
|
Sustainability model |
Certified after the fact |
Engineered in from the foundation |
|
Energy efficiency |
Industry-average PUE ~1.56 |
Designed for best-in-class PUE |
|
Power delivery |
Open-air switchyard |
In-house GIS substation, lower losses |
|
Cooling |
Air, density-limited |
Liquid (DLC, RDHx, immersion) for 100 kW+ racks |
|
Water |
Evaporative, high draw |
Closed-loop, low WUE |
|
Monitoring |
Periodic checks |
Real-time BMS + DCIM |
|
Certification |
Varies |
LEED (Chennai, proposed), Tier III+ |
|
Carbon / power |
Grid default |
Clean-power PPAs, Scope 2 focus |
Built for a Cleaner Decade
India is going to build a great many data centres this decade. The real question is whether it builds them in a way the grid and the climate can carry. Here’s the part that’s easy to miss: the constraints that look like obstacles to a green build — the strained grid, the scarce water, the slow land — are actually the forcing function for one. When you can’t brute-force capacity, you have to engineer it, and the operator compelled to be efficient will out-build the ones that never had to be. That is what sustainable by design means for us: power delivered with minimal loss through in-house GIS, liquid cooling that holds high density without wasting energy or water, real-time control that keeps PUE low in practice, and clean-power procurement that pulls Scope 2 down. Green data centres in India aren’t the slower, costlier option. On this grid, in this decade, they’re the only version of scale that actually works.
To see how a sustainable, high-density build maps to your own infrastructure plans, explore our Chennai and Mumbai campuses, or talk to our team.
Sources: IEA (global data-centre electricity, cooling energy); Uptime Institute (industry PUE, water tracking); S&P Global (India power demand, renewable targets); IBEF / Colliers (India capacity, green-certified share); Mordor Intelligence (grid connection lead times, renewable PPAs); LBNL (liquid-cooling savings, water use); World Bank (embodied carbon).