What’s Really Limiting India’s Data Centers Growth?

India crossed 1,700 megawatts of data center capacity in 2025, on the back of 56 billion dollars invested that year alone. Every announcement counts the same things: megawatts, acres, fiber routes, subsidies. Almost nothing counts heat. That is a mistake, because the ability to keep next-generation AI hardware within safe operating temperatures may become one of the defining constraints on India’s data-center buildout.

That is a mistake, because the ability to keep next-generation AI hardware within safe operating temperatures may become one of the defining constraints on India’s data-center buildout.

A modern AI accelerator such as Nvidia’s H100 draws approximately 700 watts. Next-generation AI compute modules are already pushing beyond 1,400 watts. Pack seventy-two of them into a rack, as Nvidia’s GB200 system does, and a single rack integrating GPUs, CPUs, networking and other supporting electronics can dissipate up to 132 kilowatts. The thermal challenge becomes immediately apparent.

Ten years ago, a full rack drew 5 to 10 kilowatts. Air cooling, the technology most Indian facilities were designed around, hits a practical ceiling near 40 kilowatts per rack: air cooling becomes increasingly inefficient and difficult to implement as we approach this level of heat dissipation.

The cost shows up on the electricity bill. In a conventional data center, cooling can account for 30 to 40 percent of a data center’s total energy consumption. When a chip runs hot, it throttles itself: the processor automatically reduces clock speeds to protect itself. An expensive GPU cluster running at reduced clock speed is capital sitting idle because of heat.

India adds one more variable. Summer temperatures across much of the country sit between 35 and 45 degrees Celsius for months at a time. Every degree of outside heat is capacity the cooling plant must buy back before it does any useful work. This is not a weakness of Indian engineering. It is a boundary condition, and boundary conditions decide which technologies win. It means the return on better thermal materials is higher in India than almost anywhere else.

I spent fifteen years at Bell Labs working on thermal management in semiconductor manufacturing, and three decades in academic research on heat transfer since. One lesson repeats. Before any chiller, cold plate or cooling tower can do its job, heat must first travel out of the silicon, across a series of solid interfaces, and into a fluid. Much of the thermal resistance resides within those first few millimeters, and those millimeters are governed by materials.

Liquid cooling is rapidly becoming the preferred solution for the highest-density AI systems, and rightly so. But liquid cooling does not eliminate the need for better materials. Before heat ever reaches a cold plate or coolant, it must travel from the silicon, across package structures, thermal interface materials and heat spreaders. Improving those thermal pathways lowers junction temperatures, enhances reliability and allows the cooling system itself to operate more efficiently.

This is where graphene and its family of carbon nanostructures enter the picture. Pristine graphene possesses exceptionally high in-plane thermal conductivity, several times better than copper. However, practical graphene-based materials such as multilayer graphene, graphene oxide, graphene quantum dots and graphene composites achieve more modest but still significant improvements and are increasingly being engineered for thermal management applications. Each has a place in the thermal chain, and the applications are further along than most people assume.

Switch between the four points in that chain below.

Much of the thermal resistance in a data center sits in the first few millimeters, and those millimeters are governed by materials.

None of this is merely laboratory speculation. Several of these technologies are already in commercial use, while others are advancing rapidly from laboratory research toward industrial deployment. What is missing is manufacturing scale, and a coherent feedstock strategy.

Here is the part of the story I find most consequential for India. Graphene and its relatives can be derived from coal. India holds among the largest coal reserves in the world, even as the energy transition is steadily narrowing coal’s future as a fuel. The same tonne of coal tells two different economic stories. Burned, it is worth its calorific value, once. Converted into graphene oxide or engineered graphite, or other advanced carbon materials, it becomes a feedstock for high value carbon materials that sell at many multiples of that value and feed industries that barely existed a decade ago.

Viewed this way, coal stops being a legacy liability and becomes a strategic materials resource. AI data centers could provide one of the first large-scale domestic markets for these advanced thermal materials. India is creating the demand while already possessing the raw material. What sits between them is a processing, qualification and commercialization gap, and that gap is closable.

Three steps would close it.

India’s data center ambition is real, bold and well-funded. Whether that ambition reaches its full potential will be decided not by how many megawatts get built, but on how efficiently the heat gets out. The answer lies as much in materials as in machinery, and the raw material is already beneath India’s feet.