E will eat grass or leaves, even go hungry, but we will get one of our own ... We have no other choice!` Zulfikar Ali Bhutto had famouslyremarked as he voiced this nation`s determination to build an indigenous nuclear deterrent programme that had, by the mid1960s, become an existential necessity. Some seven years later, Pakistan made it a policy imperative. That one call guaranteed the future security of the republic. Decades later, working through sheer grit and determination, Pakistan showed the world that it was more than capable of looking after itself. We stand at a very similar juncture today. The ongoing arms race in AI threatens to upend the global economic and security order, and the turning point is likely to arrive well before we will be prepared to confront it. That is, unless we once again take control of our destiny and start shaping it with our own hands today.

A large part of our annual $4.6bn in IT exports is freelancing work, call centres, and entry-level coding, which modern AI directly imperils. At the rate at which the technology is progressing, it will rapidly render many of those services and products either low-value or useless. There is also the question of what we, as an economy, must pay in terms of compute bills if our companies continue to license AI usage from foreign corporations.

As dependence on AI grows, the bill could reach hundreds of millions of dollars by the end of the decade. At most critical risk is our defence infrastructure, where drones, electronic warfare, and high-end sensors already run on AI technology, and are only going to grow more dependent on it to maintain their efficacy.

The national leadership must, therefore, make haste. In the 1970s, Pakistan was able to offer expat talent a compelling reason to return. Similar arrangements need to be made once again.

Talented computer scientists and experts from around the world must be called home, given a deadline, generous budgets, and autonomy. The systems and data centres can be placed close to where power is cheap and stable, while the research can take place where our universities are. A realistic target would be to host open-weight models at home, have researchers and technologists fine-tune them on Pakistani data and languages, run all government and defence needs within the country, and own the data pipelines as well. This would make Pakistan sovereign enough in AI without requiring a frontier-level lab or the 100-billion-dollar budgets that they need. We can build on our ambitions once the basics are in place. The Chagai test took more than two decades; the clock on AI runs much faster than that. Pakistan cannot afford to delay this project. The downside risks are too great.