TCS just pointed at India's place in the global AI infrastructure race, and it did so at a scale the country has not seen before.
On Saturday, September 5, its subsidiary HyperVault announced it would build a 1 GW AI data centre campus at Future City in Hyderabad, backed by up to Rs 70,000 crore in planned investment. Telangana's Chief Minister A Revanth Reddy set an even firmer target, asking TCS to open the facility by June 2, 2028.
This is not a single hall of servers. It is a purpose-built campus aimed at frontier AI labs and hyperscalers running high-density GPU clusters for training, inference, and advanced computing. The figures are large enough that the interesting questions are engineering, not marketing.
What the numbers actually imply
Before the rhetoric, the headline figures carry weight. Rs 70,000 crore is roughly $7.4 billion at the current dollar to rupee rate near 94. The campus sits on 264 acres secured at Future City, and its rated capacity is 1 GW from full build-out.
To put 1 GW in perspective, consider the electricity. One gigawatt sustained over a year is about 8.76 terawatt-hours, roughly the annual consumption of 800,000 US households running at full load continuously. Most modern data centres land in the 50 to 200 MW range, so a 1 GW site is several times larger than a typical hyperscale block.
The land footprint alone does not tell the whole story, because only part of 264 acres carries compute. The rest is power infrastructure, cooling loops, water treatment, and construction buffers. Still, packing a city-scale electrical load onto 264 acres is an exercise in density that most Indian industrial land cannot support today.
| Metric | Figure | What it means for builders |
|---|---|---|
| Capacity | 1 GW | ~5 to 20x a typical hyperscale data centre block |
| Investment | Rs 70,000 crore (~$7.4B) | Heavy capital on compute and cooling, light on land |
| Land | 264 acres, Future City, Hyderabad | Power and cooling dominate the footprint |
| Cooling | Liquid-cooled compute | Required for high-TDP AI accelerators |
| Water | Water-neutral design | Cools the most politically sensitive risk factor |
| Timeline | 18 to 24 months to operational | Fast for the scale; deadline set June 2, 2028 |
| Jobs | Up to 7,000 direct and indirect | Construction plus long-term operations |
Why this matters, read two ways
A data scientist looks at a 1 GW AI campus and thinks about what actually draws that current. Training and inference on frontier models are dominated by power-per-rack, not chip count. A single high-end accelerator like NVIDIA's H200 or Titanium sits near 700W, and a densely packed inference rack of dozens of those runs into tens of kilowatts per shelf. Air cooling hits a wall around 30 to 40 kW per rack. Past that, liquid cold plates or direct-to-chip cooling stop being optional. HyperVault naming liquid-cooled compute in the announcement is the tell, this campus was designed for dense GPU loads from day one, not retrofitted for them.
A software engineer reads the same plan and notices the operational cadence. HyperVault said it will develop in phases tied to customer demand and technology requirements. That is the right pattern for 2026 AI infrastructure, where model sizes and rack power densities move faster than concrete cures. Build a power and cooling block, fill it with GPUs when a hyperscaler signs, then stand up the next block. It avoids the classic mistake of building capacity you cannot sell before the next generation of chips arrives.
The water question is the real constraint
Everyone focuses on the rupee figure, but the water-neutral design principle is the more technically load-bearing choice. Cooling the hottest GPU racks consumes enormous volumes of water, and that is the factor that most often turns a great data centre site into a political problem. New projects have been delayed or reshaped over water usage, and water-positive commitments are becoming a licensing requirement rather than a goodwill gesture.
Saying water-neutral up front is an engineering commitment, and it also reads as risk management for a campus sized at 1 GW. Get the water story wrong at that scale and nothing else on the plan matters.
The timeline and the politics
TCS CEO and MD K Krithivasan framed the project as part of the company's "Infrastructure-to-Intelligence" strategy, pairing AI-ready hardware with TCS's cloud, engineering, and enterprise software businesses. That vertical integration is worth watching, it turns a data centre from a real estate bet into a distribution channel for TCS's own services.
The government side moved unusually fast. Telangana IT Minister D Sridhar Babu said the agreement grew out of a single conversation between Revanth Reddy and Tata Sons chairman N Chandrasekaran at Davos in January 2026, with officials closing the deal within months. Reddy setting a hard June 2, 2028 inauguration deadline is a political bet on delivery, and it signals how seriously the state treats this as a flagship.
The 18 to 24 month path to operational status is aggressive for a campus of this size, but plausible because the phased model lets early blocks come online while later ones are still being built.
Where this fits
India has spent years as a software services exporter. A 1 GW dedicated AI campus is the country trying to become an infrastructure exporter as well, and the first at this scale. It also tightens the grid, cooling, and supply-chain questions that other Indian projects, including ones far behind in the political pipeline, will now have to answer.
The same debate playing out in Texas, where a governor's resistance to grid hardening reportedly made that state a risky place to build new AI data centres, is exactly the gamble Telangana is backing against with this project. TCS's own move reads like a bet that government speed can win where regulation stalls.
Our read
Two things make this more than a press release. First, the liquid-cooled design at 1 GW targets the power-density wall that is now the binding constraint on AI scaling, not a nice-to-have. Second, the phased, demand-led build model matches how fast rack power and chip performance actually move, which reduces the risk of stranded capacity.
The open question is execution. Rs 70,000 crore and a 2028 deadline are commitments, and the water-neutral and grid-connection details behind those words will determine whether the campus hits its rated capacity on schedule.
Sources: The Economic Times, covering the TCS announcement The Hindu, on the June 2, 2028 deadline Swarajya, on the investment and campus design