MUMBAI, 31 AUG 2026 — Indian businesses announced ₹26.75 lakh crore of investment between April and August, with AI and data centres taking 56 per cent of it. Actual private capital spending in the first half of the financial year was ₹7.7 lakh crore.
Two numbers doing different jobs
₹26.75 lakh crore is announcements. ₹7.7 lakh crore is money spent, and it was up 67 per cent year on year, the strongest reading in over a decade.
Both numbers are correct, and they measure different things. An announcement is an intention with a press release; capital expenditure is what appears in accounts. The gap between them is normal, because announced projects are staged over years and some never proceed at all, but the two figures are being quoted in the same breath as though the larger one were the achievement.
A 56 per cent concentration is not a broad revival
The headline is a broad capital expenditure revival. The composition of the spending is narrower.
If AI and data centres account for 56 per cent of the total, then every other sector in the Indian economy — manufacturing, chemicals, textiles, automotive, pharmaceuticals — is sharing the remaining 44 per cent. A revival with that shape is one industry building at extraordinary speed while the rest of the economy invests at ordinary rates.
The distinction has consequences for employment. Broad-based capital investment signals confidence across an economy and tends to produce employment across it too. A data centre build-out is capital-intensive and famously light on permanent headcount, which means the same rupee produces a fraction of the jobs the historical relationship would imply.
The domestic share is the notable figure
86 per cent of announced investment, ₹23.01 lakh crore, is from domestic private firms.
This is a significant change, and the most durable part of the trend. India's previous infrastructure cycles leaned heavily on public spending and foreign capital, both of which are conditional on things India does not control. Domestic corporate balance sheets funding the build is a different and sturdier base, and it reflects a decade of deleveraging in Indian corporate credit.
It also changes who bears the risk. If AI demand moderates, the losses sit with Indian companies and Indian lenders rather than being distributed to foreign investors, which is a concentration worth naming while the cycle is still going up.
A twenty-year tax holiday, to 2047
The Union Budget for 2026-27 grants a twenty-year tax holiday, running to 2047, for foreign cloud providers serving global customers from Indian facilities.
Read the conditions. It is for foreign providers, and it is for serving global customers, which means the target is export-oriented capacity rather than domestic cloud supply. India is offering two decades of tax relief to host other countries' workloads.
The logic is defensible. Export capacity still consumes Indian power, land and labour, and it puts India on hyperscaler maps. The cost is a fiscal benefit deferred for twenty years, while power and water consumption starts immediately and the workloads are foreign. Whether that trade is good depends on figures nobody has published.
Renewables are in the headline and small in the data
The framing pairs data centres with renewables as twin drivers. The figures do not support equal billing.
Renewables drew ₹25,000 crore. Against announcements of ₹26.75 lakh crore that is under one per cent of the total, and it sits behind nuclear and conventional power in the sector ranking. Renewables are in the sentence because they are adjacent to the data centre story, not because they are comparable in size.
The connection between the two runs the other way. Data centres need firm power around the clock, which is what conventional and nuclear generation supply and what solar and wind do not without storage. A build-out of this scale therefore pulls harder on the sources that can guarantee output than on the ones that cannot, and the sector ordering in the data reflects exactly that.
The capacity number worth holding onto
Installed colocation capacity is around 1.5 GW, expected to exceed 6.5 GW by 2030.
A projected fourfold increase to 6.5 GW by 2030 is the concrete version of the announcement figure, and the one to check against later. It is also where the announcement-versus-delivery discipline bites, because data centre capacity has a specific failure mode: the constraint is rarely capital and almost always grid connection.
We reported that Malaysia's 6GW pipeline is announcements rather than connections, with interconnection the binding step. India's stated 2030 target is the same order of magnitude, in a country with a larger grid and a longer history of transmission constraints. The announced pipeline and the connected capacity will diverge, and the second number is the one that describes reality.
What this means from Southeast Asia
India competing directly for the hyperscaler capacity that Johor, Batam and Jakarta have been winning is the regional consequence, and the tax holiday is aimed squarely at it.
Malaysia's advantage has been land, power and proximity to Singapore. India's is scale, an enormous domestic engineering workforce and now a twenty-year tax exemption specifically designed to attract export-serving capacity. Those compete for the same investment committee decision.
For the region, this raises the bar rather than removing the opportunity. A hyperscaler deciding where to put the next gigawatt now has a credible fourth option with a fiscal incentive attached, and the ASEAN markets that have been competing on land and electricity will find that incentives alone stop being the differentiator. Grid connection speed is what remains, and it is the thing none of these markets has solved.
View author profile → · Editorial policy
