Have you every thought what would be the biggest challenge for the growth of AI in the next 10 years?
Most people would say better models, better algorithms, maybe even better data. I used to think the same. But the more I read and the more I think about it, the answer starts shifting in a completely different direction. The real bottleneck is not software anymore. It is infrastructure. It is where all of this intelligence actually runs.
This question actually started from a very simple thought. Why is there no massive AI data center in Pune? We have talent, we have engineers, we have startups, we have decent connectivity. Then why is it not happening here at scale? And the more I dug into it, the more I realized that building an AI data center has very little to do with where developers live. It has everything to do with physics.
An AI data center is not just a building with servers. It is closer to a factory. A factory that consumes insane amounts of electricity, produces a lot of heat, and needs to operate continuously without failure. We are talking about facilities that can consume anywhere between 50 to 150 megawatts of power. That is not a tech problem. That is a city level infrastructure problem. Now imagine trying to plug something like that into an already stressed urban grid like Pune. It just does not scale.
Then comes cooling. This is where things get even more interesting. These systems generate heat at a scale that is hard to visualize. You are basically running thousands of GPUs at full load for hours or days. Cooling them is not optional. It is existential. Traditional cooling methods rely heavily on water. We are talking about millions of liters of water being used daily in some setups. Now think about Pune again. Water shortages are already a recurring issue. So even if you solve power, water becomes a constraint.
Then comes connectivity. You cannot just build a data center in the middle of nowhere and expect it to work. It needs to be connected to high capacity fiber networks. It needs to serve users with low latency. Ideally you want your data center within a certain distance from your users so that response times stay low. This usually means you are constrained by geography again. So now you have three hard requirements stacking up against each other. Power, cooling, and connectivity.
This is where the idea of geography starts becoming the most important variable in the entire equation. Not talent. Not cost of developers. Geography. And once you start thinking like this, a pattern starts emerging.
Regions between roughly 40 to 65 degrees latitude in the northern hemisphere seem to hit a sweet spot. These regions naturally have cooler climates for most of the year. This means you can reduce your dependence on artificial cooling systems. Less cooling means less water usage and lower energy overhead. At the same time, these regions are still close enough to major population centers in North America, Europe, and parts of Asia. So latency remains manageable. You are not pushing your infrastructure too far away from your users.
Now look at the countries that fall into this band. The Nordics, parts of Canada, northern US states, Germany, UK, Ireland, even parts of Russia and northern China. This is not a coincidence. A lot of large scale data center investments are already happening here. Companies are not choosing these locations randomly. They are optimizing for physics, not for hype.
This brings me to a slightly extreme but interesting example. Greenland. At first glance it sounds ridiculous. Why would anyone build a data center there? But if you break it down, it actually makes sense on paper. Greenland has massive freshwater reserves. It has huge potential for renewable energy, especially hydropower. The climate is naturally cold which is perfect for cooling. And geographically it sits between North America and Europe which gives it a strategic advantage for connectivity.
But this is where reality kicks in. Greenland is promising, but it is not ready. Infrastructure is limited. Logistics are hard. Building and maintaining facilities there is expensive and complex. There are political and economic challenges that cannot be ignored. So while it looks like an ideal candidate from a purely physical perspective, the ecosystem required to support large scale deployment is still missing.
So the real answer is somewhere in between. You need a place that balances all of these factors. You need cheap and abundant power, preferably renewable. You need a climate that helps with cooling instead of fighting against it. You need strong fiber connectivity. You need land that is affordable and scalable. And you need geopolitical stability so that long term investments are safe.
Now coming back to India, the obvious question is whether regions like Kashmir or the Northeast could work. On paper, they tick some boxes. Cooler climates compared to the rest of India. Potential for hydropower. Lower population density in some areas. But again, the challenges are not trivial. Connectivity is still not at the level required for hyperscale data centers. Infrastructure development is uneven. There are geopolitical sensitivities in certain regions. Logistics and supply chains are harder compared to more industrialized zones.
So could they work in the future? Possibly. But right now, if you are building at scale, you would still lean towards regions with stronger existing infrastructure. Maybe parts of Himachal or Uttarakhand could evolve into something interesting over time if connectivity and power infrastructure improve significantly. But as of today, India’s data center growth is more concentrated around places like Mumbai, Hyderabad, and Chennai because they are closer to network hubs and undersea cable landing stations, even if they are not ideal from a cooling perspective.
And that brings us back to the original thought. Why not Pune? Because Pune is optimized for people, not for machines at that scale. It is great for building software, not for hosting the physical backbone of AI.
The more I think about it, the more it becomes clear that the future of AI is going to be shaped as much by geography and energy as it is by algorithms. We are entering a phase where compute is not just about chips, it is about where those chips live. And the places that understand this early will have a disproportionate advantage.
So maybe the real question is not where should you build the next AI data center. The real question is which regions are willing to redesign themselves to support this new kind of infrastructure. Because that is where the next wave of AI will actually come from.