This page owns the commercial data-centre and digital-economy story. It does not own Dholera's municipal smart-city ICT systems (networks, sensors, command-and-control for city operations). That is Chapter 15, Smart city & ICT. The two are related but different: one serves the city, the other is a business that might locate inside it.
One more thing before anything else. Every company project named on this page is currently at announced or proposed stage. Chapter 6 — Ground reality classifies the L&T Vyoma and Tillman Global data-centre leads as MoU-level and exploratory, with no Dholera-specific land allocation, approval, financial close or operating facility in evidence as of the last verification cycle. Announced stays announced until dated records say otherwise. This page explains the underlying logic of the sector; Chapter 6 tracks the project status.
21.1 Why Dholera is suitable for data centers
Strip away the press releases and a data centre buys four things:
Electricity, cheaply and reliably
Power is typically the largest operating cost of a data centre, and reliability determines whether the facility can hold its uptime guarantees. A location with planned high-capacity supply is structurally attractive.
A way to remove heat
Every kilowatt a server draws becomes heat that must be moved out. Cooler ambient air and available water make that cheaper. This is why large facilities historically cluster in cool or coastal regions.
Land at scale
Hyperscale campuses run to tens or hundreds of acres. In Mumbai or Chennai, land is scarce and costly. A greenfield site with large contiguous plots removes a hard constraint.
Low-latency fibre
Data centres must sit close to the fibre routes that connect users, cloud regions and submarine cable landing points. Latency to end users is a physical fact, not a marketing line.
Dholera's pitch on each of these comes from planning rather than accident. Large land parcels in an activation area with trunk infrastructure, a designed power system, treated and recycled water, and a coastal location on the Gulf of Khambhat. Whether the sector actually arrives is a separate question from whether the designed conditions fit its needs, and that distinction runs through this whole page.
21.2 Data-center infrastructure
It helps to know what is actually inside one of these buildings. A typical facility has:
- White space: the halls holding racks of servers and storage.
- Power train: incoming substations, transformers, switchgear, and backup generators plus uninterruptible power supplies, layered so that no single failure takes the building down.
- Cooling plant: chillers, cooling towers or liquid-cooling loops, and the air handlers that move heat away from equipment.
- Network: multiple independent fibre entries, so no single cable cut isolates the site.
- Security and monitoring: physical access control, fire suppression and continuous environmental monitoring.
The engineering standard is usually expressed in tiers, from Tier I (basic) up to Tier IV (fault-tolerant, designed for near-zero downtime). Higher tiers cost more to build and operate. Some announced Dholera projects have referenced Tier IV-class ambitions; that is a design statement by the companies, not a certification that exists today.
21.3 Power requirements
Power density is the number that separates eras of computing. Traditional enterprise halls ran at a few kilowatts per rack. AI training clusters now push tens of kilowatts per rack and beyond, which changes everything downstream: busbars, cooling, floor loading, even the electrical grid interface.
For a hyperscale campus, total demand is measured in hundreds of megawatts, sometimes gigawatts. This is why announced projects tie themselves to power figures. Any capacity number attached to a Dholera data-centre project should be read as announced capacity per project materials, not installed or sanctioned capacity. The designed power backbone behind Dholera's industrial zones, including its renewable supply from the planned energy system, is part of the appeal, but announced megawatts and delivered megawatts are different facts.
21.4 Cooling
Roughly a third or more of a conventional data centre's energy goes to cooling, so geography matters. Dholera sits on a coast, and coastal air is generally cooler and more humid-stable than inland plains, which improves the economics of air-side and evaporative cooling for part of the year. Gujarat's summers are hot, though, so no honest description can call the climate ideal year-round. What coastal siting offers is a better starting point than a hot inland location, not an exemption from heat.
For high-density AI racks, air alone stops being enough. Direct-to-chip liquid cooling, where coolant flows to cold plates on the processors, is the technology referenced in at least one announced Dholera project. Liquid cooling reduces the dependence on ambient air and shifts the constraint toward water treatment and distribution quality instead.
21.5 Water and recycled water
Evaporative cooling consumes water. Cooling towers consume water. A large campus can draw meaningful volumes, which is why water-stressed cities increasingly push data centres toward recycled sources or closed-loop designs.
This is one of Dholera's cleaner structural stories. The master-planned water system includes treatment and a recycled-water loop as part of the trunk infrastructure, so a facility can, in principle, source cooling water from the recycled stream rather than competing with drinking supply. The same logic applies to the semiconductor fab, which is why the two sectors read well together on this specific piece of infrastructure. As always, planned infrastructure must be built and commissioned before it serves anyone; the plan is real, the deliveries are staged.
21.6 Fiber connectivity
Latency is physics. Data travels through fibre at about two-thirds the speed of light, so every hundred kilometres of distance adds roughly a millisecond of round-trip delay. For many workloads that is trivial. For high-frequency trading, real-time applications and multi-site synchronisation, it is decisive.
Dholera's connectivity story rests on two legs. First, its position on the corridor between Ahmedabad and the coast, tied into national fibre backbones by planned trunk networks. Second, India's broader submarine-cable geography: most international bandwidth lands on the western and southern coasts, and Gujarat's coast is within reach of those routes. The connectivity projects page tracks what is actually built. What matters for this chapter is that a data centre site without diverse, redundant fibre entries is not a viable data centre site, whatever else it has going for it.
21.7 Hyperscale infrastructure
"Hyperscale" describes facilities built at enormous scale, usually by or for the largest cloud and internet companies: hundreds of megawatts, standardised building designs repeated across a campus, and construction paced in phases. India's hyperscale capacity has concentrated in Mumbai and Chennai, largely because of submarine-cable access, and both cities now face land scarcity and power constraints.
The case for new locations, Dholera among them, is that the binding constraints have moved. When land and grid capacity get expensive in the incumbent hubs, greenfield sites with planned infrastructure become competitive. The counter-case is ecosystem depth: the contractors, spare-parts networks, and skilled operations staff that cluster around established hubs. A greenfield site must build that human ecosystem too.
21.8 AI infrastructure
AI changed the data-centre requirement profile. Training clusters pack specialised accelerators at densities conventional halls never saw, and the power and cooling envelope shifts accordingly. A purpose-built AI campus in 2026 looks different from a colocation hall of 2015: liquid cooling, higher-voltage distribution, and much larger electrical interfaces.
India also has a policy layer here. National AI missions and data-localisation rules under the DPDP Act 2023 create demand for compute hosted on Indian soil. Announced Dholera projects lean on exactly this framing: sovereign AI, models trained on Indian infrastructure. The framing is plausible as demand. Whether it converts into built capacity is a project-by-project question, and Chapter 6 tracks it honestly: every convert I have checked so far has stopped at the announcement step.
21.9 Cloud infrastructure
Cloud regions are the consumer-facing version of the same infrastructure: a cloud provider builds a cluster of data centres in a location, and customers provision servers there. Region selection weighs latency to population centres, power costs, redundancy and policy. A new cloud region in Gujarat would serve western India with lower latency than Mumbai-based regions for some users, though Mumbai's cable access keeps it central.
Nothing on this page should be read as a cloud region announcement. The point is structural: if hyperscale campuses get built, cloud capacity is one of the plausible uses, and the same power, cooling and fibre requirements apply.
21.10 Green data centers
Data centres are energy-hungry, so their carbon story matters increasingly to both regulators and customers. "Green data centre" usually means one or more of: power sourced from renewables, high efficiency measured as power usage effectiveness (PUE), water conservation, and waste-heat reuse.
Dholera's renewable planning, anchored by the planned solar energy system, makes the renewable-power part of that list structurally available in a way that few Indian sites can claim from a standing start. Announced projects have tied themselves to solar supply explicitly. The honest caveat is the same one from the power section: renewable ambition, contracted renewable supply and delivered renewable electrons are three different milestones. PUE figures, when companies publish them for real buildings, are earned numbers; announced PUE targets are design intent.
21.11 Major announced projects
Three projects dominate the current announcement record. Each is classified 🟠 ANNOUNCED/PROPOSED in Chapter 6, last verified 17 September 2026, with no Dholera-specific land, approval, financial close or operating facility in evidence. Figures below are as stated in project materials and announcements.
| Project | Announced scope (per project materials) | Status on our record |
|---|---|---|
| Adani group data-centre ambitions (AdaniConnex JV) | Large-scale hyperscale plans referenced in group materials, alongside a broader stated data-centre investment ambition to 2035; renewable-powered compute and sovereign AI cited as focus areas | 🟠 ANNOUNCED/PROPOSED · no Dholera-specific facility evidenced; current status requires verification |
| Tillman Global Holdings | Announced around a US$10 billion scale figure and a 1 GW+ "digital complex" concept, referencing Tier IV ambitions and solar-linked power | 🟠 ANNOUNCED/PROPOSED · exploratory lead only per Ch.6; current status requires verification |
| L&T Vyoma | Announced as a ~₹25,000 crore, 250 MW AI-focused "green campus" concept with direct-to-chip liquid cooling, via an MoU signed February 2026 at the India AI Impact Summit | 🟠 ANNOUNCED/PROPOSED · MoU-level per Ch.6; current status requires verification |
Aggregate figures sometimes quoted in coverage, including totals in the lakh-crore range across all announced projects, are sums of announcements, not committed capital. We report them only with that label attached.
21.12 Digital economy ecosystem
Data centres rarely justify themselves in isolation. The ecosystem that forms around them includes:
- Colocation and interconnection providers that rent space and connect networks.
- Managed services and systems integration firms that run workloads for clients.
- Edge facilities closer to users, complementing the big campuses.
- Skills: electrical, mechanical and network engineers, plus facilities operations staff, who need training pipelines and places to live.
- Anchor demand: government cloud workloads, enterprise migration and AI development, which determine whether capacity gets filled.
Dholera's smart-city ICT planning (Chapter 15) and its industrial promotion both imagine a digital-economy layer, but an ecosystem is grown, not declared. The realistic sequence is one anchor facility operating, then suppliers and service firms following. That sequence has not started here as of the last verification.
21.13 Semiconductor + compute integration
Of everything in Dholera's plan, this is the pairing I find most interesting: the one between the semiconductor chapter and this one. A fab makes chips; data centres are among their largest buyers. Locating compute near fabrication is not common anywhere in the world yet, and no Dholera project currently commits to it, so what follows is design logic, not a project claim.
The pairing works on paper because both sectors draw on the same infrastructure bones: very large, very clean power; enormous quantities of treated and recycled water; large contiguous land parcels; and fibre. A fab and an AI campus both stress a local grid in ways that ordinary industry does not, and a master-planned grid sized for both from the start is cheaper than retrofitting one later. The energy chapter covers that backbone.
There is also a supply-chain reading. Chips designed and fabricated in India, then trained on and hosted in Indian data centres, would close a loop that currently runs through Taiwan, the United States and Singapore. Policy frameworks, including semiconductor mission incentives and AI compute initiatives, point in that direction. Whether the loop closes in Dholera specifically depends on the fab delivering, the announced data centres converting into buildings, and the water and power systems arriving on schedule. Three conditions, none yet met, all trackable.