Silicon Valley’s earnings calls are already answering the most important question about artificial intelligence: where the money ends up. There’s an undeniable sense of triumph in India’s tech ecosystem right now, but we might be celebrating a digital victory we haven’t actually won.

If you look across the country, the physical transformation is staggering. We are investing massive capital in hyper-scale data centre parks in Navi Mumbai. Legacy IT giants are weaving AI into enterprise networks and digital public infrastructure. The government has stepped in to subsidise tens of thousands of GPUs (graphics processing units) for startups. We are doing everything right on the hardware front.

But beneath the ribbon-cutting and the applause lies an uncomfortable economic reality: India’s much-hyped AI boom is directly funding Silicon Valley.

In effect, we have signed up for a “Token Tax”. Every single time a developer in Bengaluru uses an AI coding assistant, or an e-commerce app personalises a shopping feed, or a local bank automates a customer query in Hindi, a micro-payment leaves the country. We are pouring concrete, laying the fibre-optic cables, and buying the servers. But the actual intelligence — the cognitive engine running on top of all that infrastructure — remains foreign-owned.

We are building the toll roads, but someone else is collecting the toll.

You can already see this imbalance playing out on global balance sheets. While we celebrate our domestic infrastructure push, HSBC projects that the world’s leading tech giants will pull in $2.8 trillion in revenue by 2026. A rapidly growing share of that wealth is being extracted from countries exactly like ours — markets that are adopting AI at breakneck speed, but failing to own the underlying models. The global cloud monopolies aren’t threatened by our large-scale hardware investments; they are banking on them.

The perpetual utility bill

To understand how we got trapped here, you have to look at how the fundamental business model of software has shifted.

Back in the 2000s, Indian IT firms operated in a world of ownership. You bought a database licence, installed it on your servers, and built an empire of high-margin services on top of it. It was a one-time capital expense. You owned the tool. That era is over. Today, AI is metered. You rent it by the token.

Indian companies are doing the hardest, most unglamorous work in the digital economy: acquiring price-sensitive users, navigating complex local regulations, and structuring messy domestic data. Yet the highest-margin layer of the entire value chain — the API fee for the intelligence itself — flows immediately outward. India is no longer just exporting software services. We are importing intelligence, one API call at a time.

Math doesn’t work at scale

The usual pushback I hear from tech leaders is predictable: Why not just rent? Renting a frontier model from a US tech giant is infinitely faster, highly reliable, and cheaper than spending billions to train our own. Why not just focus on building great local apps and let California handle the heavy lifting?

The answer is simple: that model mathematically cannot survive Indian economics.

Our market is defined by enormous volume but incredibly tight margins. The average revenue per user (ARPU) here is a fraction of what companies earn in the West. You simply cannot serve hundreds of millions of users while paying a dollar-denominated, per-token tax to Silicon Valley for every interaction.

Consider an agri-tech startup offering AI-driven crop advice to farmers in Marathi for a ₹50 monthly subscription. If every complex question that farmer asks requires a round-trip to a foreign AI model costing a few cents, the math collapses. The more successful the product becomes, the more money it bleeds.

This is why so many brilliant Indian AI startups are currently stuck in “pilot purgatory”. The product works, the local demand is real, but scaling it destroys their balance sheet. That isn’t a healthy tech ecosystem. It is a dependency loop.

A UPI-style playbook for AI

The solution isn’t some blindly idealistic push to build everything domestically at any cost. Trying to create a bloated, state-run ‘National AI Reserve’ will likely just give us a slow, bureaucratic system that the private sector ignores.

But we have navigated this exact kind of trap before. Look at digital payments. A decade ago, India flat-out refused to hand over its payment infrastructure to global incumbents like Visa and Mastercard. We built the Unified Payments Interface (UPI). India built the digital rails, retained ownership of the core system, and let private companies innovate on top of it. We desperately need a similar playbook for AI.

The government’s role shouldn’t be to build the models, but to underwrite the immense compute required for a consortium of private Indian firms, researchers, and startups to develop competitive, sovereign models.

Simultaneously, corporate India needs to wake up. Not every internal workflow or document-processing task requires pinging a massive, trillion-parameter global model. Enterprises need to pivot towards Small Language Models (SLMs) that can be trained on their own proprietary data and run safely on local servers. This keeps costs down, protects privacy, and entirely cuts the external gatekeepers out of the transaction.

In this new era, data gravity — keeping your data and computation local — is the only real defence against algorithmic rent.

The invisible deficit

We obsessively track our trade deficit when it comes to oil, gold, and electronics. We know exactly what physical imports cost our economy. But we are completely blind to our “Compute Deficit”.

India spends roughly $130 billion a year importing oil. If our digital adoption continues on this path, the cost of relying on foreign AI APIs could eventually rival that scale. Millions of daily API calls are already draining capital out of the country, yet this massive outflow is practically invisible in our national accounting.

For three decades, India exported human talent to power the global digital economy. But in the generative AI era, selling human effort is a depreciating asset. Intelligence itself is becoming the commodity, and right now, we don’t own it.

Globally, other nations are waking up. France is heavily backing Mistral to ensure European sovereignty. The UAE is pouring billions into the Falcon models. China has insulated its entire ecosystem. India risks becoming the only major digital economy that defaults to pure import for its cognitive engine.

India now faces a clear choice. We can continue subsidising hardware and exporting the profits, settling for scale without ownership. Or, we can do the harder work of shifting from adoption to control. Because the arithmetic is unforgiving. If nothing changes, by 2030, India’s most expensive import won’t be Arabian oil or Chinese electronics. It will be intelligence. And we won’t just be buying it — we will be renting our own future.

The writer is a physicist at the University of North Carolina at Chapel Hill and a columnist on AI, infrastructure, and global systems

For three decades, India exported human talent to power the global digital economy. But in the generative AI era, selling human effort is a depreciating asset

Published on April 15, 2026