Compute is the new railway

I got my first computer when I was 10: a humble Pentium III with 512 MB RAM, magic in a box for my younger self!

Fast forward, I upgraded to an i5 with 4 GB RAM, a loyal companion for years. Then an i5 with 8 GB, followed by an i7 with 16 GB. And today? I'm working on an Apple M3, marketed with "neural" capabilities and delivering a few teraflops. That is trillions of calculations every second, on a laptop.

Impressive? Sure. But here's the real jaw-dropper: compare that to the GPU clusters that train frontier AI models, which scale into exaflops. That's like comparing a bicycle to a spacecraft.

The revolution beneath the surface

What I didn't fully contextualise back then was the massive revolution quietly shaping our world.

• Subsea cables and mobile networks connecting the globe, laying the foundation of today's internet.
• Chip design innovations, with billions of transistors crammed onto a single chip, making compute better and better year after year.
• The rise of compute power as a currency, where controlling it means controlling influence and global leadership.

This isn't just about better compute. It's about shaping the future. Today, nations aren't just competing for economic power, they're racing for AI compute.

From $500 billion mega-initiatives like Stargate to chip export restrictions, the stakes have never been higher. The message is clear: sovereign AI must become a top national priority for India.

Just as railways and roads built the industrial age, data, compute and connectivity rails will define the AI economy.

Why rails, and not trains

History is a good teacher here. The railways didn't just move goods. They decided which towns grew, which ports mattered and which industries were possible at all. Whoever laid the track shaped the economy that ran on it, for a century.

AI works the same way. The model gets the headlines, but the model is the train. The rails are underneath: the chips, the data centres, the energy that powers them, the networks that connect them, and the data that feeds them. You can rent a seat on someone else's train. You cannot build a national economy on rented seats.

I spent six years at Bosch building AI and IoT businesses, and one lesson kept repeating itself. The smartest product means little if it sits on infrastructure you don't control. Pricing, access, latency, even whether you are allowed to use it, all of it is decided upstream.

The five rails

It helps to be specific about what "the rails" actually are, because each has a different owner and a different bottleneck.

• Chips. The design and manufacture of advanced processors is concentrated in a handful of companies and countries. Access can be restricted by export rules as well as by price.
• Data centres. Buildings full of compute need land, permits, cooling and years of construction. Capacity cannot be conjured up when demand arrives.
• Energy. AI is electricity turned into intelligence. Countries with reliable, affordable and increasingly clean power will host more of it.
• Connectivity. Fibre, subsea cables and mobile networks decide how quickly compute reaches users and businesses.
• Data. Models reflect what they are trained on. Without datasets in a country's own languages, industries and contexts, the intelligence they produce will always be borrowed.

A serious national strategy has to work on all five at once. Strength in one cannot make up for weakness in the others.

The wrapper question

Next time someone talks about why some nation's firms are building "AI wrappers," pause and ask: what powers it beneath the surface?

Wrappers are not bad. Many great businesses start as a smart layer on top of someone else's platform. But a country is not a startup. If every layer below the application belongs to someone else, then so does most of the value, and most of the leverage.

India has done this before. We built rails for money, identity and data that the world now studies as digital public infrastructure. That is our proof that we can lay track at national scale when we decide to. Government programmes to make shared GPU capacity available to startups and researchers are an encouraging start in the same spirit.

What this means for business leaders

This is not only a question for governments. Every company using AI is making infrastructure decisions, whether it realises it or not.

• Know your dependencies. Which models, clouds and chips does your AI roadmap rely on, and what happens if prices, terms or access change?
• Keep your options open. Architectures that can switch between models and providers are worth a little extra effort.
• Treat your data as an asset. Proprietary data about your customers, operations and products is the one rail you fully own.
• Plan for energy. For companies running AI at scale, power costs and availability are becoming strategic, not just operational.

Are we building the rails?

The real question is whether we are building the infrastructure to keep up with this revolution. Nations investing in these rails today are shaping tomorrow's possibilities.

Three questions I would put to any leader, in government or in business:

• Compute: do we have access to it on our own terms, or only on someone else's?
• Data: are we creating the datasets that reflect our languages, our industries and our people?
• Talent: are we training people to build the rails, or only to ride on them?

Without the right infrastructure, the AI race might already be over. With it, India could be laying the tracks others run on.