AI infrastructure capex surge
Investors are shifting from flashy AI apps to the underlying hardware and facilities that run them, treating chips and data‑centre capacity as the cleaner way to capture AI demand. India is a clear theatre for that shift, with hyperscalers and domestic groups driving a fast data‑centre build‑out that changes where companies place compute and why electricity economics matter. TSMC is also planning massive US investment to meet processor demand while advanced-node capacity is being reserved by the largest buyers, signalling tighter supply for everyone else. (techaimag.com) (thehindubusinessline.com) (intellectia.ai)
Investors are putting more of their artificial intelligence money into chips, servers and data centres than into the apps built on top of them. (thehindubusinessline.com) In India, data-centre capacity rose from about 0.375 gigawatts in 2020 to around 1.5 gigawatts in 2025, and hyperscalers such as Amazon Web Services, Microsoft, Google and Oracle already account for more than half of installed information-technology load. (thehindubusinessline.com) Jones Lang LaSalle said India had 1,123 megawatts of data-centre inventory in the first half of 2025, net take-up of 97.9 megawatts, and a vacancy rate of 4.3%, with total capacity projected to reach 2,073 megawatts by the end of 2027. (jll.com) The business is now measured as much by electricity as by floor space, because artificial-intelligence workloads need far more power and cooling than older computing jobs. The Hindu BusinessLine reported that large language models can require 10 to 100 times the computing power of traditional applications. (thehindubusinessline.com) That is changing where companies build. Mumbai and Chennai hold much of India’s installed co-location capacity because they sit near submarine cable landing stations, while operators are also looking at Tier 1 and Tier 2 cities as local grids tighten. (thehindubusinessline.com 1) (thehindubusinessline.com 2) Power costs are moving to the center of the investment case. The Hindu BusinessLine reported in February 2026 that data centres already use up to 15% of local grid capacity in some areas, and about 30% to 35% of current Indian capacity is backed by renewable power purchase agreements. (thehindubusinessline.com) Colliers said India’s data-centre market stood at 1,263 megawatts in April 2025 and could cross 4,500 megawatts by 2030, drawing $20 billion to $25 billion of investment over five to six years. Mumbai had 41% of capacity, followed by Chennai at 23% and Delhi National Capital Region at 14%. (colliers.com) The chip side is tightening too. Taiwan Semiconductor Manufacturing Company said on March 4, 2025 that it would add $100 billion to its United States expansion, taking planned spending in Arizona to $165 billion with three new fabrication plants, two advanced-packaging facilities and a research-and-development center. (tsmc.com) Taiwan Semiconductor Manufacturing Company said the Arizona project is now planned as six wafer fabs, two advanced-packaging facilities and a research-and-development team center, and called it the largest foreign direct investment in a greenfield project in United States history. (tsmc.com) Big technology companies are matching that build-out with bigger budgets. CNBC reported on February 4, 2026 that Alphabet’s capital spending will go to artificial-intelligence compute capacity for Google DeepMind and cloud demand, while Bloomberg reported on February 6, 2026 that Amazon, Alphabet, Meta and Microsoft together could spend about $650 billion in 2026. (cnbc.com) (bloomberg.com) The result is a market where the cleaner bet, for many investors, is no longer the next artificial-intelligence app but the physical capacity underneath it: the chip lines, the server halls and the power contracts that keep them running. (thehindubusinessline.com) (tsmc.com)
