Centamil designs the 1 MW white shell: a factory-built, liquid-cooled data-centre module that rejects its heat to dry air. No cooling towers, no make-up water, and a design PUE below 1.15.
Most AI capacity is cooled by chillers and evaporative towers. A site like that runs at a PUE of about 1.35, so it draws about 35% more power than its computers use. It also evaporates about 1.8 litres of water for every kilowatt-hour of computing.
Telangana alone expects 8–10 GW of data-centre capacity around Hyderabad, against more than 300 MW operating today. At that scale, every point of PUE and every litre per kilowatt-hour becomes a permanent demand on the grid and the reservoirs.
1.8 L of water evaporated
0 L of make-up water for cooling
Basis: a PUE of 1.35 and 1.8 litres per kWh are typical figures for chiller-cooled and evaporative-cooled sites. White-shell figures are design targets; each drop above is 0.1 litre. Capacity figures are from Government of Telangana statements in 2026.
Every watt a chip draws leaves it as heat. In a white shell that heat leaves in liquid, travels a sealed loop, and ends in dry air.
Direct-to-chip cold plates take the heat off the processors, and coolant distribution units move it on. Racks run at up to 130 kW each. One shell holds seven or eight.
The facility loop is closed. It is filled once and topped up by less than 1% a year. There are no cooling towers and no evaporative pads.
Each shell has its own dry heat-rejection unit. Fans draw ambient air across the coils and the heat leaves as warm air. Nothing evaporates, so cooling needs no make-up water. With no compressors on the main cooling path, the design PUE is below 1.15.
Batteries sit on the shell’s DC bus, in line with the 800 V DC direction of AI racks. They carry short outages, firm the site, can support the grid, and cut generator run hours.
A shell is factory-built, delivered and commissioned in months. Customers add their own compute racks. Power and water are metered at every unit, and we will publish measured performance from our first installations.
Design targets. Illustration, not to scale.
We supply the white shell: enclosure, power distribution, liquid cooling, dry heat rejection and storage integration. You bring the compute.
These are design targets. We will publish measured performance from our first installations.
Set the size of the build. Each square below is one megawatt; each block of 25 is a campus like the one we propose in Telangana.
MW of AI capacity, built as white shells
Illustrative, against a chiller-cooled baseline of PUE 1.35 and evaporative water use of 1.8 litres per kWh of IT energy, with the white shell at its design PUE of 1.15 and at full IT load all year.
Proposed, subject to approvals
We are proposing a 25 MW reference campus at Bharat Future City, Telangana: 25 white shells, each with its own dry heat-rejection unit, sharing power, storage, backup and fibre.
Phase 1, January to June 2027
We fit our dry heat-rejection outer loop to an existing data centre in Hyderabad, then measure and publish the water and power it saves through the summer of 2027.
Phase 2, fourth quarter of 2027
The first commercial white shell at Bharat Future City: liquid-cooled, with no make-up water and a design PUE below 1.15.
Phase 3, 2028
We add DC-coupled grid-scale storage, and move to 100% dry heat rejection all year, with no trim chillers.
Phase 4, 2028 to 2030
The site grows in blocks as compute partners commit, with its own substation, battery storage, standby generation and an operations centre with a visitor gallery.
Phase 5, from 2028
A Centamil production facility in Telangana assembles white shells with local vendors, for our own campus and for clients in India and abroad. Many of those sites will be small, a megawatt or a few, close to the people who use them.
Indicative layout, not to scale. Subject to land allotment and approvals.
The campus is planned to host AI compute for our partners and to double as a working showroom. Buyers considering a white shell can visit, watch units run through a Telangana summer, and read the published PUE and water data.
Each shell is built with local vendors, first for our own campus and then for clients in India and abroad. We also plan a technician programme in liquid cooling and data-centre operations with local institutions.
If you already operate a data centre
Our dry heat-rejection outer loop also retrofits into existing data centres, to cut their water use. It is where our own programme starts in 2027.
If you are a regulated organisation
Banks, insurers, capital-market firms and government departments increasingly have to keep data and AI workloads on infrastructure they control. In India the drivers are RBI data-localisation rules, the DPDP Act, CERT-In directions and sovereign AI programmes. That means many small, secure data centres of 0.5–5 MW, close to their users. A water-free, modular 1 MW unit is built for that.
Centamil Ltd is an IP licensing and design house in Masdar City Free Zone, Abu Dhabi. We design the system and hold its IP: the dry heat-rejection loop and the modular 1 MW unit, with direct-to-chip cold plates developed with Hayagreeva Inc.
We build with partners. Local fabricators make the shells, operators and compute providers run them, and financiers fund them, through a project company formed in each market.
Founder and CEO
A chemical and embedded-systems engineer who has transferred 28 technologies to international companies, including Unilever, L’Oréal and Philips Healthcare. Forbes 30 Under 30 honoree, National Science Medal winner and recipient of a President of India award.
Centamil also develops AI-embedded microreactors for process industries. Read about them at centamil.ai.
7–9 December 2026, Bharat Future City, Hyderabad