Pakistan’s new sovereign AI play

‘Vibe coding’ is the new AI trend: software is moving from written to requested. That shift will jolt Pakistan’s IT success story because a large slice of our export base still depends on outsourced, task-based work.

When applications can be generated from natural-language prompts, the economics flips: routine build work becomes cheaper, while the ability to deploy trusted systems – at scale, safely and reliably – becomes the differentiator.

Put numbers on it. Pakistan’s ICT services exports reached $3.2 billion in FY24, with $350 million in freelancing remittances – around 15 per cent of the ICT total. And this is not a marginal sector. Pakistan is rated as a financially attractive IT outsourcing destination on Kearney’s Global Services Location Index, with a domestic base of 600,000+ English-speaking IT professionals.

Pakistan also has greater latent technological strength than its broader economy suggests. Research by economist Cesar Hidalgo uses the geography of open-source contributions on GitHub to infer national capabilities, indicating that Pakistan’s software sophistication exceeds what its manufacturing profile alone would predict.

The capability is real, but so is the risk of AI automation. Much of what Pakistan exports falls within the first wave of automatable work – basic programming, digital media design, data cleaning – as AI compresses the labour-arbitrage edge that made outsourcing viable.

Can Pakistan use this moment to upgrade its model: from selling hours to selling outcomes; from service provision to building systems others can rely on. The path exists, but it demands more than ‘more coding’. It requires sovereignty, as properly defined.

That is why ‘sovereign AI’ matters – but only if we define it correctly. On February 9, 2026, the government of Pakistan convened the Indus AI Summit and issued the Islamabad AI Declaration, setting principles and priorities for Sovereign AI and ‘governing intelligence at scale’, with an emphasis on public value, accountability, data trust and practical delivery. The direction is right: the next phase of competition won’t be won by the flashiest demo. It will be won by whoever can deploy AI in the real economy with dependable outcomes.

Yet sovereign AI is often mis-sold as a procurement wish list – national foundation models, national clouds, national everything. Pakistan, like most middle powers, does not need to own the full stack to be sovereign, and it cannot. Frontier models are inherently global; the US and China dominate the frontier due to their inherent scale and concentration of talent. The compute race is being driven by hyperscalers whose FY2026 capital expenditure is forecast at $600 billion – well beyond the reach of most emerging markets.

The more strategic target is cognitive sovereignty: can Pakistan deploy applications in ways that serve national interests and reflect societal values, even if parts of the stack are external?

Ownership is a balance-sheet concept. Sovereignty is an operating concept – classifying which critical workloads are and which are not, setting enforceable rules for data stewardship and building an architecture that avoids lock-in to any single provider. If you don’t define your sovereign workloads, someone else will.

This is also where Pakistan must resist the prestige trap. The near-term wins are not at the frontier-model layer; they are at the application layer – where domain knowledge, regulation, trust and distribution decide success. Agriculture yields optimisation, health triage and education personalisation: these are not consolation prizes. They are the fastest route to productivity at home – and to the bigger prize: exportable applications, systems Pakistan builds that others adopt and integrate into their economies; not one-off tasks bought on a platform.

Finally, sovereignty does not require silo-ism. No country solves compute, capital and talent simultaneously on its own. Selective cooperation can strengthen sovereignty. A pragmatic regional ‘triangle’ is worth testing through joint projects: Pakistan as a talent-rich economy; Saudi Arabia as a capital-and-energy platform for compute; Turkiye as entrepreneurial execution with experience in global markets. The kick-start can be joint project funding, shared sandboxes in regulated sectors and structured talent exchanges that build teams capable of shipping.

In today’s AI race, headlines frame the world as a US-China binary. For most middle powers, that framing is paralysing: it tempts them to chase ownership of a stack they cannot replicate, while neglecting the layer where sustainable power can be built. The credible response is not self-sufficiency but agency – build relevant applications, align with peers to pool comparative advantages, and be ruthless in pursuing what ‘sovereignty’ must mean in practice.

Pakistan’s freelancing and outsourcing exports are both a success story and a warning light. When AI can deliver ‘good enough’ output instantly, the market will pay less for routine effort and more for accountable systems. Pakistan’s choice is straightforward: defend the old model until it disappears or ascend the AI value chain – building trust and exportable applications while we still have momentum.

Malik Ahmad Jalal is based in Karachi and Ussal Sahbaz in Istanbul. They are Harvard graduates and work in technology policy and venture investing.