TCS Plans a $7.4 Billion AI Data Center: 1 GW of Compute, HyperVault, Telangana, and India’s AI Infrastructure Expansion

HyperVault, a subsidiary of Tata Consultancy Services, has secured 264 acres in Hyderabad for a new AI data center campus with planned capacity of up to 1 gigawatt. HyperVault and its partners expect to invest up to ₹70,000 crore, or roughly $7.4 billion at contemporaneous exchange rates, making the project one of the largest AI-infrastructure commitments announced in India.

The campus is being designed specifically for frontier AI companies and hyperscalers, with high-density GPU deployments for model training, inference, and advanced computing workloads. TCS says the development will proceed in phases, tied to customer demand and changing technology requirements rather than being delivered as a single 1 GW block from day one.

The technical architecture is as important as the headline capacity. HyperVault is emphasizing direct-to-chip liquid cooling, high rack density, large power blocks, resilient network connectivity, green-energy integration, and water-neutral design principles—all of which address the power and thermal characteristics of modern accelerator clusters.

The project also sits inside TCS’ broader Infrastructure-to-Intelligence strategy. HyperVault is backed by the wider Tata ecosystem and TPG, while TCS has separately disclosed AI-infrastructure partnerships with OpenAI and AMD. Those relationships provide important context for the platform, but the September 5 Hyderabad announcement does not assign the entire campus to any single customer.

THE HYDERABAD CAMPUS: 264 ACRES, UP TO 1 GW, AND ₹70,000 CRORE.

The announced specifications establish the physical scale, target workloads, and build philosophy of the project before detailed phase-by-phase engineering data has been released.

A 1 GW campus is not simply a very large conventional colocation facility. At that scale, power sourcing, grid interconnection, electrical distribution, cooling, fiber capacity, equipment procurement, and phased customer occupancy become core design constraints rather than secondary operating details.

........

Parameter	Announced detail	Technical implication
Developer	HyperVault AI Data Center Limited, a TCS subsidiary	Places the project inside TCS’ dedicated AI-infrastructure business rather than its traditional IT-services delivery model
Location and land	Hyderabad, Telangana; 264 acres secured	Provides room for a campus-scale build with multiple data halls, power systems, cooling infrastructure, substations, network facilities, and phased expansion
Planned capacity	Up to 1 GW	Supports very large accelerator clusters and multiple anchor workloads if power and customer demand are delivered in phases
Expected investment	Up to ₹70,000 crore, roughly $7.4 billion	Signals that the project includes not only buildings but also substantial electrical, cooling, networking, and operational infrastructure
Target customers	Frontier AI companies and hyperscalers	Design requirements are driven by dense training and inference clusters rather than general-purpose enterprise racks
Target workloads	AI training, inference, and advanced computing	Requires sustained high power density, high-bandwidth networking, thermal management, and reliable power delivery
Build model	Phased according to customer demand and technology requirements	Reduces the risk of overbuilding capacity before contracted workloads, while allowing later phases to adopt newer hardware generations
Cooling and sustainability	Liquid-cooled infrastructure; green-energy and water-neutral design principles	Addresses accelerator thermal density while placing energy and water engineering at the center of the campus design
Employment	Several thousand direct and indirect jobs expected	Extends the economic impact into construction, power, networking, engineering, operations, and supply-chain services

........

The phrase “up to 1 GW” should be read as a full-build target, not as immediately commissioned capacity. TCS has not disclosed a complete construction timetable, exact phase sizes, GPU count, rack-density figure, PUE target, contracted megawatts, or the precise electrical load available on day one.

That distinction is material because AI data centers are increasingly contracted in large power blocks. TCS said earlier in 2026 that discussions with hyperscalers and frontier-model companies were converging around anchor workloads in the 100–200 MW range per customer. A campus built in that environment is likely to be planned around discrete customer and power modules rather than one monolithic facility.

The 264-acre footprint also gives HyperVault room to separate compute halls from the supporting infrastructure required at this scale. Substations, backup systems, cooling plants, water-treatment or water-reuse systems, network meet-me rooms, security zones, logistics, and future power expansion can consume substantial land around the compute buildings themselves.

··········

WHY THIS IS AN AI DATA CENTER, NOT A CONVENTIONAL CLOUD FACILITY.

The design requirements come from accelerator density and cluster behavior: modern AI systems concentrate far more power, heat, and east-west network traffic into each rack than traditional enterprise computing.

The first difference is power density. Large GPU and accelerator systems can place tens or hundreds of kilowatts into a single rack, depending on platform configuration. That changes electrical distribution from the utility feed all the way to busways, rack power delivery, redundancy design, and maintenance planning. A campus may have nominal gigawatt-scale capacity, but usable compute capacity depends on how efficiently that power is converted into continuously available rack-level power.

The second difference is cooling. HyperVault explicitly describes direct-to-chip liquid cooling as part of its platform. Liquid systems move heat away from processors more efficiently than air-only approaches at very high rack densities, but they add pumps, coolant distribution units, heat exchangers, leak-management systems, water-quality controls, and maintenance procedures that must be engineered into the facility from the beginning.

The third difference is networking. Distributed training requires thousands of accelerators to exchange parameters and intermediate data with very low latency and high bandwidth. A facility can have sufficient power and GPUs yet still underperform if the interconnect fabric, optical links, topology, or congestion management cannot sustain the communication pattern of the workload. The data-center network therefore becomes part of the AI system rather than a generic transport layer.

The fourth difference is deployment cadence. AI hardware generations change quickly, and rack-level designs can shift materially between accelerator platforms. HyperVault’s phased approach gives it the option to build later halls around newer power, cooling, and interconnect requirements instead of locking the entire 1 GW target into a single generation of infrastructure.

The fifth difference is utilization risk. Training clusters are capital intensive, while inference workloads may have very different traffic patterns, latency requirements, and hardware economics. Building only against contracted or visible demand can reduce stranded-capacity risk, especially when the infrastructure itself must be optimized for specific rack-scale systems.

Sustainability constraints are also operational constraints. TCS says the Hyderabad campus will use green energy and water-neutral design principles. At gigawatt scale, the availability, timing, and reliability of low-carbon electricity can affect the pace of build-out, while water-neutral design requires a credible balance between cooling architecture, local water use, reuse, recycling, and replenishment strategies.

··········

HYPERVAULT’S BROADER COMPUTE STRATEGY: OPENAI, AMD, TPG, AND THE TATA ECOSYSTEM.

The Hyderabad project is one part of a broader attempt by TCS to move down the AI stack from services and integration into owned or controlled compute infrastructure.

HyperVault was created as TCS’ dedicated AI-data-center business and is being built with strategic support from TPG and the wider Tata ecosystem. TCS describes the model as Infrastructure-to-Intelligence: physical AI infrastructure is intended to connect with its cloud, engineering, enterprise-transformation, and AI-services businesses rather than operate as an isolated real-estate asset.

........

Element	Confirmed scope	Relationship to the Hyderabad announcement
HyperVault + TPG	TPG is a strategic partner in HyperVault; TCS has positioned the business for GW-scale AI infrastructure in India	Provides capital and infrastructure context for the broader HyperVault platform
OpenAI	TCS previously disclosed a multi-year HyperVault agreement for 100 MW of initial AI infrastructure in India, with an option to scale to 1 GW	Important customer context, but the September 5 release does not state that the full Hyderabad campus is dedicated to OpenAI
AMD	TCS and AMD have said they are co-developing rack-scale AI infrastructure based on AMD’s Helios platform, including a blueprint supporting up to 200 MW	Shows that HyperVault is being designed to accommodate specific next-generation accelerator architectures
Tata ecosystem	TCS has identified Tata Power, Tata Projects, and Tata Communications among the ecosystem participants supporting power, construction, connectivity, and delivery	Creates an integrated domestic supply chain around the data-center build
Industrial partners	TCS has also cited companies including GE, Honeywell, ABB, and Siemens in the broader ecosystem around power, cooling, controls, network, and security	Supports specialized infrastructure layers needed for high-density AI campuses
Hyderabad campus	264 acres, up to 1 GW, phased development for frontier AI companies and hyperscalers	Provides a large physical site through which those infrastructure and customer relationships can be deployed over time

........

The OpenAI relationship requires careful interpretation. TCS has separately stated that HyperVault and OpenAI agreed to develop AI infrastructure in India with 100 MW in the initial phase and an option to scale to 1 GW. TCS’ FY2026 annual report described OpenAI as an anchor customer for an initial 100 MW phase. The September 5 Hyderabad press release, however, does not say that the entire Hyderabad site is an OpenAI campus, so the two announcements should not be collapsed into a single claim.

AMD adds a different layer to the strategy. TCS and AMD have been working on a rack-scale infrastructure design based on the Helios platform and an AI-ready data-center blueprint supporting up to 200 MW. That is relevant because AI campuses increasingly need to be engineered around specific accelerator, rack, cooling, power, and networking combinations rather than around generic server specifications.

TCS is also using capabilities that already exist elsewhere in the Tata Group. Power generation and procurement, large-scale engineering and construction, fiber connectivity, systems integration, and enterprise relationships can all reduce the number of external interfaces that HyperVault must coordinate. The strategic question is whether that integration translates into faster delivery, lower execution risk, and competitive economics once customers compare HyperVault with global hyperscalers and specialist AI-infrastructure providers.

··········

WHAT THE PROJECT CHANGES FOR INDIA’S AI INFRASTRUCTURE MARKET.

The Hyderabad campus is a bet that access to large blocks of AI compute will become a strategic infrastructure layer in India, not simply a cloud service purchased from capacity located elsewhere.

For frontier-model developers and hyperscalers, local gigawatt-scale capacity can reduce dependence on overseas infrastructure, support data-residency and sovereign-compute requirements, and create new options for locating training and inference closer to Indian users and enterprise customers. For TCS, it also creates a route into a part of the AI value chain where demand is measured in megawatts, accelerator clusters, and long-term capacity commitments rather than consulting headcount alone.

The economic spillover extends beyond servers. TCS expects the project to stimulate activity across power, cooling, networking, construction, engineering, and operations. A fully developed AI campus needs substations, transmission upgrades, switchgear, transformers, generators or other backup systems, liquid-cooling equipment, optical networking, security systems, and specialist maintenance. These layers can become local industrial capabilities if procurement and engineering remain anchored in the region.

The constraints are equally significant. A nominal 1 GW target requires dependable electricity at a scale comparable to major industrial loads. Grid connection schedules, renewable-power availability, transformer and switchgear lead times, accelerator supply, financing costs, and customer commitments can all determine whether the final campus reaches its stated capacity and how quickly each phase becomes economically productive.

Customer concentration is another risk. Large AI campuses are often built around a small number of anchor tenants because a single frontier-model company can consume 100 MW or more. That makes contracted demand valuable, but it also means utilization can become sensitive to the expansion plans, hardware choices, and capital budgets of a relatively small number of customers. HyperVault’s phased construction model is therefore not merely a construction preference; it is a mechanism for matching capital deployment to real demand.

The technical success of the campus will ultimately be judged by delivered rack power, thermal performance, network efficiency, uptime, deployment speed, and usable accelerator capacity rather than by the 1 GW headline alone. If HyperVault can convert the announced land, power, cooling, partner ecosystem, and customer pipeline into repeatable 100–200 MW-class deployments, Hyderabad could become one of India’s most important physical nodes for frontier AI infrastructure.

For TCS, the project also changes the company’s exposure to the AI cycle. Instead of participating only through software, cloud integration, consulting, and enterprise transformation, HyperVault places TCS directly inside the capital-intensive infrastructure layer that determines how quickly advanced models can be trained and served. The opportunity is larger, but so are the execution, power, utilization, hardware-obsolescence, and financing risks.

The announced ₹70,000 crore ceiling should therefore be read as the scale of the ambition rather than a guarantee of immediate expenditure. The more informative milestones will be the size of the first energized phase, contracted customers, delivered megawatts, cooling design, accelerator platforms, renewable-power arrangements, and the speed with which later phases move from land and engineering plans into active compute.

·····

FOLLOW US FOR MORE.

·····

·····

DATA STUDIOS

·····

[datastudios.org]