By Shree Harsha, Senior Sales Director – Growth Industries, Dassault Systèmes
India is building AI infrastructure at a pace few could have imagined five years ago. Major hyperscalers are committing billions. New data centres are coming up across Mumbai, Hyderabad, Chennai and Pune. The government has backed this with policy, subsidies and ambition.
And yet, a very real question sits at the heart of all this progress.
Can India power this AI boom without running out of power and water to do it?
The Numbers Are Not Comfortable
India’s data centre capacity has already crossed 1.5 GW, with further growth expected through 2026. Adding capacity is no longer just a construction problem, it depends heavily on access to reliable electricity. Based on under-construction and planned projects, India’s IT load is projected to increase to 6.5–17 GW by 2030.
A single 100-megawatt data centre uses approximately 2 cubic metres of water per day for cooling alone. When you add India’s AI and cloud computing ambitions together, the sector could consume nearly 37.5 billion litres of water annually.
And the pressure does not come only from water. Modern GPU clusters generate far more heat than traditional servers. A single rack can now exceed 40 kW, with next-generation systems pushing beyond 50 kW. Traditional air cooling was not designed for this. Liquid cooling is becoming necessary. But liquid cooling needs engineering precision. It needs infrastructure that is designed right from day one. Getting this wrong is expensive. Getting it wrong at scale is irreversible.
India Is Water-Stressed. And That Matters Here
Chennai, Hyderabad, Mumbai, Bengaluru, these are India’s data centre hubs. According to the World Resources Institute, all these hubs are classified as ‘Water Stressed’ or ‘Water Scarce’. Chennai faced ‘Day Zero’ in 2019 yet remains a top destination for server farms due to its undersea cable landings.
During the 2024 Bengaluru water crisis, tech parks were asked to cut usage. As AI adoption grows, this conflict between corporate cooling and municipal drinking water will only sharpen. These are not distant problems. They are happening in the same cities where AI infrastructure is being built.
Every tech giant in the world is interested in building AI data centres in India. Late last year, Microsoft, Amazon and Google committed tens of billions of dollars to doing so. The Adani Group pledged USD 100 billion, bringing the combined first wave to USD 167.5 billion. The question is not whether to build. The question is how to build it right.
Design Is Where the Battle Is Won
I have spent a large part of my career helping companies make complex decisions about infrastructure, operations and digital transformation. One thing I have learned: the most expensive mistakes are always the ones made during design. A data centre that is built without simulating its thermal loads, airflow, cooling loops and power distribution will always underperform. When it underperforms at the scale we are talking about hundreds of megawatts of AI compute the consequences go beyond inefficiency. They become waste. Of energy. Of water. Of capital.
This is exactly why the industry needs a different approach. Not a better guess. A better blueprint, in fact a virtual twin.
A virtual twin is a real-time digital replica of a physical data centre system, enabling the simulation and optimisation of system performance before making costly real-world changes. Stakeholders can leverage the virtual twin experience to model and simulate compute systems, as well as their energy supply and cooling infrastructure. This way, they can maximise energy efficiency through improved heat, liquid, cooling water and airflow management.
For an AI data centre, this means you can model exactly how heat will move through a high-density GPU rack. You can test different liquid cooling configurations without touching a pipe. You can simulate what happens to power draw when workloads spike at 3 AM.
For data centres, virtual twins can help with power and cooling optimisation, capacity planning for AI workloads, failure prediction and more. For India, where we are building fast and at scale, this is more of a necessity.
Data centres’ consumption of all available electricity and cooling water on Earth could increase from 3% today to 8% in 2030. Most facilities only optimise PUE and WUE after they are built. A virtual twin lets you optimise both during design months before commissioning.
This Is Not Just About Efficiency, but Sovereignty
India wants to be an AI leader. That ambition requires data sovereignty the ability to keep sensitive computational workloads within our borders, on infrastructure we own and control.
Sovereignty is becoming a key differentiator the ability to guarantee that high-value operational data stays where it should. When you open an AI factory in the region, it needs to run, and being able to guarantee sovereignty becomes critical.
But sovereign AI infrastructure cannot be built the same way traditional data centres were. It needs to handle the intensity of modern AI workloads while staying within the water and energy constraints of a water-stressed, grid-dependent country. That is a hard engineering problem. And it needs an engineering solution, not a policy document. Virtual twins give us the architectural blueprint to solve this. They let us design sovereign AI factories that are also sustainable ones. That balance is achievable. But only if we design for it from the start.
3 key business impact of Virtual Twins are as below:
Accelerate Time-to-Revenue: Deploy AI infrastructure faster with synchronized design and deployment workflows.
Improve Tokens-per-Watt: Optimize energy efficiency across the AI Factory.
Maximize Uptime: Reduce integration issues and improve infrastructure resilience
The Call to Action Is Simple
To data centre developers and hyperscalers investing in India: bring simulation into your design process. Make the virtual twin your first build, not your last resort.
To policymakers and regulators: mandate WUE reporting alongside PUE. Create standards that reward sustainable design, not just fast deployment.
To the technology ecosystem: this is a design and engineering challenge. With virtual twins, you have a way to simulate tomorrow’s systems before they are built de-risking decisions and accelerating impact.
The tools exist. The methodology is proven. We need the will to use them from day one. Model-Based Systems Engineering (MBSE) approach help organizations accelerate deployment, reduce risk, and optimize AI Factory performance from day one.
India has the ambition, the market and the talent to build AI infrastructure that sets a global standard.
Not just in scale. In sustainability.
Let us build it that way.