Portfolio record · Enriched · Verified 22 Oct 2026
TCS HyperVault — 1 GW AI Data Centre ProgrammeTata Consultancy Services with TPG · Hyderabad, Visakhapatnam, Pune
Record PRJ-013National — multi-site programmeTier A — procurement-relevant14 public sources cross-checked
Programme — not additive264 acres securedOpenAI at 100 MWCommitted capital ≠ announced ceiling
1GW
Programme design
100MW
OpenAI anchor
₹18,000cr
Committed equity
₹70,000cr
Build ceiling
264ac
Hyderabad land
$1bn
Capex per 150 MW
TCS's HyperVault is the most capital-intensive AI data-centre programme in India: a 1 GW design, a ₹70,000 crore (US$7.4 bn) build ceiling, 264 acres secured in Hyderabad, land acquired in Visakhapatnam and Pune, and OpenAI as the first customer at 100 MW with an option to scale to a gigawatt. But only ₹18,000 crore — US$2 bn — is actually committed, split between TCS and TPG. The remaining ₹52,000 crore arrives only if tenants do. Every number below should be read in that order: what is funded, then what is announced.
1At a glance
Record ID
PRJ-013 — Gridvik India Data Centre Registry
Record
TCS HyperVault 1 GW AI data-centre programme — a portfolio programme, not a single site
Entity
HyperVault, the AI data-centre business of Tata Consultancy Services, with TPG as a partner investor
Programme design
Up to 1 GW of AI-ready capacity, over five to seven years
Announced build ceiling
Up to ₹70,000 crore (about US$7.4 bn), contingent on customer demand — not a commitment
Committed capital
About ₹18,000 crore (US$2 bn) of equity from TCS and TPG. TPG committed up to ₹8,820 crore for a 27.5–49% stake; TCS retains 51%. Announced 20 November 2025
First customer
OpenAI — 100 MW initial commitment with an option to 1 GW, announced 19 February 2026 at the India AI Impact Summit, under OpenAI's Stargate programme
Land secured
264 acres in Hyderabad (Future City), announced 5 September 2026. Land also acquired in Visakhapatnam and Pune, reported October 2026
Unit economics
TCS states about US$1 bn of capex per 150 MW — roughly US$6.7 m per MW, reflecting liquid-cooled GPU infrastructure
Child records
PRJ-014 — the OpenAI customer capacity. PRJ-063 — the Hyderabad Future City campus. Neither is additive to this programme total
Gridvik tier
A — procurement-relevant. Largest single AI-infrastructure buying programme in India, at pre-tender stage
2Funded versus announced — the only reconciliation that matters
Two numbers circulate for HyperVault: ₹18,000 crore and ₹70,000 crore. They differ by a factor of four and they are both correct. They answer different questions.
Figure 1Build cost at TCS's own stated rate of about US$1 bn per 150 MW, tranche by tranche, against the announced ₹70,000 crore ceiling. The gap of roughly 11% at the top end is headroom, not committed capital.
Two numbers, two meanings
Figure
What it is
Contingent on
₹18,000 crore
Equity actually committed by TCS and TPG, of which TPG's share is up to ₹8,820 crore for 27.5–49%
Signed, 20 Nov 2025
₹70,000 crore
The announced ceiling for the Hyderabad campus build-out including partners — 'up to', explicitly contingent on demand
Tenant demand
1 GW
Campus design capacity for Hyderabad. At about $1 bn per 150 MW this implies roughly $6.67 bn of capex before the announced headroom
Tenant demand
100 MW
OpenAI's initial commitment — about 10% of campus design capacity, and roughly two thirds of a single 150 MW tranche
Signed, 19 Feb 2026
Gridvik read — the funding gap is the story, not the headline
₹70,000 crore is a ceiling, not a cheque. The money that exists is ₹18,000 crore, and the demand that exists is 100 MW.
That is not scepticism, it is arithmetic. 100 MW of signed demand against 1 GW of design capacity means about 90% of this campus has no named tenant. The programme is credible because OpenAI's option runs to 1 GW and TCS can fund the first tranche from committed capital — but the build will be paced to leases, not to the announcement.
For a vendor, that changes the question from 'how much will be bought' to 'which tranche is funded'. Tranche one is roughly 150 MW at about US$1 bn. That is the business case worth building a quote around.
3Unit economics — why AI capacity costs what it does
TCS's stated rate
About US$1 bn of capex per 150 MW of HyperVault buildout — roughly US$6.7 m per MW
Why that is high
Liquid-cooled GPU infrastructure. Direct-to-chip cooling, higher power density per rack, CDU plant, and heavier electrical distribution all cost more per MW than conventional air-cooled enterprise colocation
Comparison
Conventional Indian hyperscale colocation has historically been built for materially less per MW than this figure. The premium is the AI specification, not the land or the shell
Anchor workload size
TCS states that engagements with hyperscalers and frontier AI model companies are converging on anchor workloads of 100–200 MW per customer
Technology blueprint
TCS and AMD are developing an AI data-centre blueprint on AMD's Helios rack-scale architecture, supporting deployments up to 200 MW
What this means commercially
At US$6.7 m per MW, this is the highest equipment spend per MW available to an Indian vendor in the market today. A 150 MW tranche carries about US$1 bn of procurement in it — more than many operators spend in a decade
What is not disclosed
Split between shell, power, cooling, IT fit-out and compute. Land cost for Hyderabad. Per-tranche timing beyond phase-one-by-demand
4The OpenAI anchor
What is actually agreed
On 19 February 2026, at the India AI Impact Summit, OpenAI and the Tata Group announced a partnership under which OpenAI becomes the first customer of TCS's HyperVault business, beginning with 100 MW of AI-ready capacity with the potential to scale to 1 GW over time, as part of OpenAI's Stargate initiative.
The deal is explicitly about data residency and sovereignty — OpenAI said the infrastructure would let its models run within India, cutting response times and supporting compliance needs including government workloads. It sits alongside an enterprise agreement under which ChatGPT Enterprise rolls out across the Tata workforce, beginning with hundreds of thousands of TCS employees, and TCS standardises AI-native development on OpenAI's Codex tools.
This is the strongest demand signal on any Indian AI data-centre programme, and it is still only 10% of the campus.
Initial commitment
100 MW
Option
To 1 GW over time
Programme
OpenAI Stargate — the non-US build-out
Legal status
Multi-year strategic partnership, signed and announced by both parties
Enterprise component
ChatGPT Enterprise across the Tata workforce, Codex for TCS engineering — commercial software, not capacity
Other tenants
None named. The stated target market is frontier AI companies and hyperscalers, with anchor workloads of 100–200 MW
What is not disclosed
Pricing, term, exclusivity, whether the 1 GW option is at fixed terms, and which campus the first 100 MW lands on
5Site pipeline
Where the capacity is going
Site
State
Status
Detail
Hyderabad
Telangana
Land secured
264 acres at Future City, announced 5 September 2026. Campus design up to 1 GW. Recorded separately as PRJ-063
Visakhapatnam
Andhra Pradesh
Land acquired
Reported October 2026 as land acquired for OpenAI data centres. No acreage or capacity published
Pune
Maharashtra
Land acquired
Reported October 2026. No acreage or capacity published
Programme total
—
1 GW
The programme figure. Individual sites are not additive to it
The additivity trap, stated plainly
HyperVault's 1 GW is a programme figure. The Hyderabad 264-acre campus is a site figure and is a child of this record. The OpenAI 100 MW is a customer figure. These three overlap almost completely.
Adding them — 1,000 + 1,000 + 100 — would produce a number 110% too high. Gridvik flags the programme as non-additive for exactly this reason, and holds Hyderabad and OpenAI as separate child records. If a competitor quotes you 2.1 GW of Tata AI capacity, that is how they got it.
6Structure and governance
Vehicle
HyperVault — TCS's AI data-centre business, held as a separate entity with external equity
TPG investment
Announced 20 November 2025. TPG to invest up to ₹8,820 crore (about US$1 bn) for a final shareholding of 27.5% to 49%. TCS retains 51%
Funding vehicle for TPG
TPG Rise Climate and its Global South Initiative, a private-equity strategy launched in partnership with ALTÉRRA
Combined commitment
Up to ₹18,000 crore (about US$2 bn) from TCS and TPG combined, funded through a mix of equity and debt, in tranches over several years
Positioning
Sovereign, enterprise-grade AI infrastructure — targeting hyperscalers, AI companies, deep-tech firms, the Indian government and large enterprises
Why TCS is doing this
TCS is an IT services company buying into physical infrastructure because its clients — including OpenAI — need sovereign, domestic compute. The services business is the demand channel, which is unusual and is the programme's main advantage
Why TPG is doing this
A climate-and-growth fund taking a large minority in India's AI compute build-out, at a moment when global capital is otherwise cautious on data-centre equity
State support
The Hyderabad campus was announced in partnership with the Telangana state government, which is a material de-risking factor for land, power and approvals
What is not disclosed
Debt quantum and lenders, per-site capex, the TPG entry price implied by the shareholding range, and any government anchor tenant
7Timeline
TCS announces TPG's investment of up to ₹8,820 crore for 27.5–49% of HyperVault, with combined commitments of up to ₹18,000 crore. TCS retains 51%.
OpenAI becomes HyperVault's first customer: 100 MW initial, option to 1 GW, under Stargate. Announced at the India AI Impact Summit alongside a group-wide ChatGPT Enterprise and Codex agreement.
The partnership is restated in TCS's Q4 FY26 results — 100 MW initial phase, option to 1 GW.
HyperVault announces 264 acres in Hyderabad at Future City, with an announced investment ceiling of up to ₹70,000 crore (about US$7.4 bn) for a campus of up to 1 GW, in phases paced to demand.
Land acquired in Visakhapatnam and Pune for OpenAI data centres.
Programme delivery window. No public schedule of which tranche is built when, beyond 'according to customer demand'.
8Procurement
Read this before the table
This is the largest pre-tender AI-infrastructure buying programme in India. Almost nothing is awarded yet, which is precisely why it matters — but it is also early, and being early means relationships and specifications are still being formed. Everything below is a Gridvik estimate.
Procurement window and scope
Gridvik assessment · 22 Oct 2026
Order window
Now through 2031, tranche by tranche. Tranche one is roughly 150 MW at about US$1 bn of capex, anchored by OpenAI's 100 MW. Because the build is explicitly paced to customer demand, each tranche is a discrete buying event rather than a continuous programme — which means a vendor needs to be positioned before each one, not once at the start.
Procurement heat
High, and unusually early. Land secured in three states, capital committed, anchor tenant signed, and no published EPC award. This is the rare window where a specification can still be influenced — a liquid-cooling vendor with a documented reference design has more leverage here than on any operational campus.
Who buys
HyperVault, with TCS corporate governance and TPG as an investor with board-level visibility. Expect a formal, documented process with lender and investor scrutiny, closer to a global hyperscaler than to an Indian developer
DG / gensets (indicative)
At 1 GW of AI load with N+1 redundancy, a full build implies roughly 800–950 gensets of 2,000–2,250 kVA class across the programme, or roughly 120–145 units for the first 150 MW tranche. Note that AI campus backup practice increasingly favours larger unit sizes
UPS (indicative)
High-density AI halls are moving toward distributed and rack-level power with battery at the rack. This may reduce centralised UPS content per MW relative to a conventional build — a genuine specification risk for UPS vendors
Cooling (indicative)
This is the defining category of the programme. The unit cost of US$6.7 m per MW is an AI-infrastructure cost, and the dominant driver is liquid cooling. That implies direct-to-chip cold plates, CDU plant, high-flow manifolds, leak detection, and rack densities far above Indian colocation norms. Liquid-cooling vendors with Indian service capability have almost no competition in this segment at present
Categories in scope
HV substations and switchgear · DG and gensets · distributed power and rack-level battery · CDUs · cold plates and manifolds · high-density racks · rear-door heat exchangers · chilled water plant and dry coolers · busway and high-current distribution · structured cabling and high-speed optics · fire detection and suppression · leak detection · BMS/DCIM · water treatment · renewables and storage
Decision role
Public decision roles — in GridFeed
Stage-change history
Dated change log — in GridFeed
All estimates are labelled as estimates. No tender dates claimed, no private contact data published.
Programme size, capital commitments, TPG terms and shareholding range, OpenAI commitment and option, Hyderabad acreage and announced ceiling, Visakhapatnam and Pune land acquisition, TCS's stated unit cost, AMD blueprint, workload size
What is derived
The procurement section, the implied 90% unleased capacity, and the tranche ladder on Figure 1
Capacity figure used
1 GW as programme design; 100 MW as signed demand. Both are carried, labelled, and never added to the site or customer records
Additivity rule
Non-additive. This is a portfolio programme. The Hyderabad campus (PRJ-063) and the OpenAI capacity (PRJ-014) sit inside it and must never be summed with it
Cost basis
The US$1 bn per 150 MW figure is TCS's own stated assumption, sourced from its disclosure. The tranche ladder is arithmetic on that rate, not a forecast
Currency
Reported in both INR and USD by the company. Gridvik uses the fixed quarterly reference rate of ₹88.40 per USD, which gives about $7.9 bn for ₹70,000 crore and about $2.0 bn for ₹18,000 crore. Company and press figures state $7.4 bn and $2 bn respectively
10Limits, conflicts and open questions
Committed versus announced
The central caution. ₹18,000 crore is committed; ₹70,000 crore is a ceiling contingent on demand
Unleased capacity
About 90% of the 1 GW design has no named tenant. Only 100 MW is signed
No EPC award
No engineering, procurement or construction contractor has been announced for any HyperVault campus. This is the largest unplaced package in Indian data centres
Tranche schedule
Not published. The build is paced to demand, so no year-by-year capacity is knowable from public sources
Shareholding
The TPG stake is a range, 27.5–49%. The implied entry valuation spans nearly a factor of two, so no reliable equity value can be derived
FX conflict
Press coverage states ₹70,000 crore as about $7.4 bn and ₹18,000 crore as about $2 bn. At the ₹88.40 reference rate those are $7.9 bn and $2.0 bn. The gap is exchange-rate vintage, not error
Tenant concentration
One customer, one programme. If OpenAI's build-out slips, the pacing of this campus slips with it
Child-record handling
PRJ-063 (Hyderabad campus) and PRJ-014 (OpenAI capacity) are indexed separately and are not additive to this record
What we will not claim
No per-site capex, no tender dates, no equipment brands, and no view on whether 1 GW will be built
11Sources
This record was compiled entirely from publicly available sources, cross-checked against one another.
TCS press release, TPG investment (Nov 2025)·TPG newsroom·TCS Q4 FY26 results·TCS HyperVault press release, Hyderabad campus·Tata Sons newsroom·OpenAI announcement·TechCrunch·Reuters·Moneycontrol·Converge Digest·DatacenterDynamics·Datacentre Magazine·DQ India·Analytics India Magazine·Blackridge Research
Gridvik does not publish source URLs on record pages; citations are retained in the internal source register and are available on request.
What this page is, and what it is not
This is a public-source record of a funded AI-infrastructure programme at pre-tender stage. Every capital, capacity and tenant statement is attributed to TCS, TPG, OpenAI or the trade press.
The procurement section is clearly labelled as estimation. No tender insider information, no private contact data, no operator-paid placement. Corrections are free and published with a date.
