India plans data‑center supercycle

Data centers are becoming one of India’s biggest infrastructure stories — not just a tech story. AI is the reason. A normal cloud build mostly needs floor space, connectivity, and predictable power. An AI build needs all of that, but denser — more electricity per rack, more cooling, more backup, more fiber, and faster construction. That is why India’s data-center expansion is starting to look less like real estate and more like an industrial capex cycle. JLL’s H1 2025 snapshot showed 1,123 MW of live capacity, 97.9 MW of net absorption in six months, and vacancy down to 4.3%, which is tight enough to keep builders moving. (jll.com) ### Why does AI change the buildout? Because AI racks are hungrier than ordinary enterprise servers. The expensive part is no longer just the white space where machines sit. The expensive part is everything wrapped around them — substations, transformers, switchgear, UPS, batteries, generators, chillers, liquid-cooling systems, and the building design needed to keep a(jll.com) clearly: electrical systems alone now account for roughly 40% to 45% of total data-center capex. (plindia.com) ### What do the numbers say? They say this is already large and getting larger fast. JLL expects India’s capacity to reach 2,073 MW by the end of 2027, up 85% from the H1 2025 base, requiring about 10.7 million square feet and roughly $6.3 billion of capital. Prabhudas Lilladher goes further on the medium-term infrastructure angle — it sees India’s data-center capacity growing from abo(plindia.com)pex opportunity. (jll.com) ### Where is the squeeze? Power first. Fiber second. Land and permits right behind them. Mumbai still holds 54% of India’s installed capacity because it has the best combination of fiber links, business demand, and cable-landing access. But that concentration also shows the problem — operators want to spread out, yet new sites only work if they can secure reliable ele(jll.com)s, EPC firms, and transmission equipment suppliers, not just colocation landlords. (jll.com) ### Why does fiber matter so much? Because AI demand does not stop at the data-center gate. It spills into metro networks, long-haul links, enterprise connectivity, and tower backhaul. India still has relatively low fiberization — around 35% to 40% by one industry estimate in late 2025, versus 80% plus in many markets. That gap matters because more AI traffic means mo(jll.com)AI adoption are all pushing in the same direction. (telecom.economictimes.indiatimes.com) ### Where do HFCL and STL fit? Mostly as picks-and-shovels suppliers. STL has been explicit that it wants more revenue from data-center and enterprise segments and is building broader product suites around that demand. HFCL has been pitching advanced fiber technologi(telecom.economictimes.indiatimes.com)They are the companies trying to sell the pipes and links that make the campuses usable. (stl.tech) ### Who is actually building the AI side? A mix of global and domestic operators. Yotta has been especially aggressive in tying Indian sovereign AI ambitions to local GPU capacity, including its NVIDIA-linked Shakti Cloud platform. Princeton Digital Group has also been expanding AI-ready capacity in Mumbai and Chennai, explicitly marketing those campuses a(stl.tech)— AI readiness is becoming a sales feature, not a niche add-on. (shakticloud.ai) ### So why call it a supercycle? Because the spend is stacking. One AI order triggers demand for land, power equipment, cooling, construction, backup systems, fiber, and cloud capacity at the same time. That is more like a mini industrial boom than a normal tech refresh. The bottom line is simple: India is not just adding more data centers. It is rebuilding the physical systems around computation — and th(shakticloud.ai)ower-and-network vendors as server-room operators.