The supercomputer India borrowed from Japan

America switched off an AI model for 19 days. India’s answer is a supercomputer in Japan — and a bet that indispensability beats sovereignty

Let me tell you about the least glamorous sentence in Indian tech policy this month, and why it might matter more than any product launch you will read about this year.

In the first week of July, on the sidelines of the 16th India-Japan Annual Summit, the two Governments signed a Joint Statement on Artificial Intelligence. Buried in the paperwork was a memorandum between Japan’s industry ministry and India’s IT ministry, under which Japan will let Indian researchers use a supercomputer operated by its National Institute of Advanced Industrial Science and Technology. In exchange, India will help ease Japan’s chronic shortage of software engineers. Tokyo has also said it wants to bring 500 highly skilled Indian AI professionals to Japan by 2030. Read quickly, it sounds like the diplomatic equivalent of neighbours lending each other kitchen appliances. You have a mixer-grinder, I have a pressure cooker, let us manage.

Read slowly, it is something else entirely. It is India quietly admitting what kind of AI power it is going to be — and making a bet that most of the world’s loudest voices think is wrong.

The thing nobody says out loud

Here is the uncomfortable arithmetic underneath all of India’s AI ambition. We are a country with more than two million AI-skilled professionals, a technology services industry that industry body Nasscom believes could capture an additional 300 to 400 billion dollars in AI-related business by 2030, and one of the largest pools of software talent on Earth. And we do not have a single frontier AI model of our own.

The chatbots in your phone, the coding assistants in Bengaluru’s offices, the models being wired into banks and hospitals and, increasingly, Government services — the most capable of them are built in San Francisco, Seattle and, lately, Hangzhou. India is the world’s great AI user, not yet its maker. For years, this felt like a comfortable arrangement. Software has always flowed across borders. Why build the expensive thing when you can rent it?

Then June happened.

For nineteen days last month, the American Government ordered one of its leading AI companies to cut off foreign access to its most advanced model, citing national security. If you were an Indian developer or company that had built your workflow on that model, you woke up one morning and a tool you paid for had simply vanished — not because of a business decision, but because of a memo written in Washington. Access was restored in early July. The lesson was not.

The lesson was this: when you rent your intelligence from someone else, the landlord can change the locks.

Two bad options and a third one

So what is a country in India’s position supposed to do? The debate in Delhi’s policy circles has mostly ping-ponged between two answers, both unsatisfying.

Option one: build everything ourselves. Sovereign models, sovereign chips, sovereign data centres, the works. It is a stirring speech. It is also staggeringly expensive. Training a frontier model now consumes computing power measured in numbers so large they sound invented, and the American companies leading the race are spending more on data centres in a year than India’s entire semiconductor mission — a Rs 76,000 crore programme - has committed in total. India’s data centre capacity stands at around 1,280 megawatts today. That is expected to grow four or five times by 2030, which sounds impressive until you learn that single American AI campuses are being planned at a gigawatt each.

Option two: keep renting and hope for the best. After June, this option should require no further comment.

The Japan pact points at a third answer, and it is the most Indian answer imaginable: be so useful to so many people that nobody can afford to lock you out.

Look at what the deal actually trades. Japan has world-class hardware, precision manufacturing, serious compute infrastructure — and a demographic crisis that has left it desperately short of the software engineers needed to use any of it. India has the exact inverse: an ocean of talent and a shortage of the expensive machinery that talent needs. Japanese officials have spent this year openly describing India as their key partner in the Global South, pairing Indian software scale with Japanese industrial depth. The supercomputer memorandum is that theory made concrete: our people, their machine.

Notice what India is not doing here. It is not pretending it can outspend Silicon Valley. It is deepening what economists would call backward linkages — securing reliable access to the ingredients of frontier AI (compute, chips, infrastructure) from multiple partners - while building forward linkages for the things India genuinely can make: applications, services, and AI that works in 22 languages for a billion people. The February AI summit in Delhi, the first global AI gathering hosted in the Global South, pushed the same idea with its Trusted AI Commons platform: if you cannot own the frontier, own the friendships.

Is renting sovereignty a bug or a feature?

The obvious criticism writes itself, and it deserves a fair hearing. A borrowed supercomputer is still borrowed. Five hundred engineers going to Japan is, from another angle, five hundred engineers leaving India. Diversifying your dependence across Washington and Tokyo is still dependence; you have merely acquired two landlords instead of one.

All true. And yet.

Consider what dependence actually looks like in each direction. Japan needs Indian talent so badly that its industrial future partly hinges on it. That is not the relationship of a landlord and a tenant. It is closer to a partnership where each side holds something the other cannot easily replace — which is, historically, the only kind of technology relationship that survives a geopolitical bad mood. The June episode showed what one-sided dependence costs. The Japan deal is an attempt to make sure India’s next dependencies are mutual ones.

There is also a quieter truth here that the sovereignty hawks skip past. The countries that “won” previous technology waves were not always the inventors. South Korea did not invent the semiconductor; it became indispensable to making them. India did not invent the software industry; it became the place the software industry could not function without. Indispensability is a form of sovereignty. It just doesn’t photograph as well as a rocket launch.

What this means for you

If you are reading this in Kolkata or Kanpur rather than a ministry office, here is the honest translation.

In the short term, very little changes. Your chatbot will still be American, your phone’s chips still Taiwanese, and the AI features creeping into your banking app still built on rented foreign intelligence.

But over the next few years, watch three things. Watch whether Indian researchers actually get meaningful time on that Japanese machine, or whether the memorandum gathers dust the way memoranda do. Watch whether the 500 engineers going to Japan come back — talent that circulates builds a country; talent that only departs builds someone else’s. And watch whether the next time a foreign Government flips a switch on an AI model, Indian businesses shrug instead of panic. That shrug, when it comes, will be the real measure of this strategy.

The loudest voices in AI insist there are only two kinds of countries: those that build the frontier and those that beg at it. India just placed a wager that there is a third kind — the country that makes itself impossible to leave out. It is an unfashionable bet, made in unfashionable paperwork, with no keynote and no applause.

It might also be the shrewdest thing Indian technology policy has done in years. Ask me again in 2030 — preferably over a video call that we can both still trust.

The writer is a physicist at the University of North Carolina at Chapel Hill and a columnist on AI, infrastructure and global systems; Views presented are personal.