business · 2026-09-06
TCS's ₹70K Cr AI Bet Isn't TCS's Money
Photo: Metropolitan Transportation Authority from United States of America / Wikimedia (CC BY 2.0)
TCS is naming a ₹70,000 crore ceiling for its Hyderabad AI campus, but its own math for building 1 GW elsewhere costs far less, because most of this bill is meant to come from future investors and customers, not TCS's own cash.
How is the ₹70,000 crore actually being funded?
TCS is not writing this cheque alone. It has already brought in private equity firm TPG, which is putting up to ₹8,820 crore into HyperVault and can end up owning 27.5-49% of the business, with the rest coming from debt rather than TCS's own cash. TCS itself said the goal was to "reduce its capital outlay" and boost shareholder returns, so most of the ₹70,000 crore is expected to come from outside investors and from customer revenue once the data centres start operating, not from TCS's balance sheet.
Why is TCS willing to give up equity and control over HyperVault?
TCS built its core business capital-light, earning a 51% ROE and over 80% ROIC in FY25 without heavy asset ownership. Data centres are the opposite: capital-heavy, slow-return infrastructure. Taking TPG's cash and debt instead of its own keeps TCS's balance sheet light while still letting it capture services revenue from AI infrastructure it doesn't fully fund. (e23)
What has to actually happen for this AI bet to pay off?
TCS's own math sets the bar: every 150 MW needs $1 billion, meaning 1 GW needs roughly $6.5-7 billion, built over 5-7 years, with revenue only starting 18-24 months in. So the number to watch is committed occupancy — tenants like OpenAI signing beyond the first 100 MW — since empty capacity means investors and lenders, not customers, are left carrying the load.
Who are these future investors TCS is counting on?
Two kinds. First, financial investors: private equity firm TPG has already committed up to ₹8,820 crore for a 27.5-49% stake in HyperVault, alongside debt, so TCS itself puts in less cash. Second, paying customers: OpenAI is lined up as the first tenant, starting with 100 MW, meaning tenant contracts, not TCS's balance sheet, are expected to fund most of the build.
Why is TCS willing to give up equity and control over HyperVault?
TCS built its core business capital-light, earning a 51% ROE and over 80% ROIC in FY25 without heavy asset ownership. Data centres are the opposite: capital-heavy, slow-return infrastructure. Taking TPG's cash and debt instead of its own keeps TCS's balance sheet light while still letting it capture services revenue from AI infrastructure it doesn't fully fund. (e23)
What has to actually happen for this AI bet to pay off?
TCS's own math sets the bar: every 150 MW needs $1 billion, meaning 1 GW needs roughly $6.5-7 billion, built over 5-7 years, with revenue only starting 18-24 months in. So the number to watch is committed occupancy — tenants like OpenAI signing beyond the first 100 MW — since empty capacity means investors and lenders, not customers, are left carrying the load.
Source: thehindu.com