GUL Ahmed Energy Group recently announced plans to build Pakistan’s largest Tier III data centre. Why is a textile group investing in AI infrastructure? Interests follow economic opportunity, and AI is potentially the biggest economic opportunity in recent times. Data centres are AI’s infrastructural layer, and that layer runs on power. Global power demand is projected to more than double by 2030, from roughly 415 TWh to nearly 945 TWh. Countries able to supply reliable, competitively priced power at scale will host the infrastructure underpinning the next wave of economic value.

Pakistan, as of July-March FY 2026, has 49,651MW of installed capacity but converts only a fraction into usable electricity. Fuel constraints, transmission bottlenecks and rigid power purchase agreements keep much of it idle, while consumers pay capacity charges for power that never reaches them. This is both a fiscal and strategic problem. Grid reform is the obvious fix, but it is too slow to wait for. The faster path is to bring compute to where the electricity already is. Data centres collocated with generation and stabilised with battery storage bypass transmission bottlenecks and monetise capacity already paid for but unused. This model needs no grid reform, only regulatory changes permitting direct contracts between generators and off-grid facilities.

Pakistan has earmarked surplus capacity for AI and blockchain infrastructure. Sky47, which runs the 8.5MW Karakoram-01 in Islamabad, is developing a 5MW facility at Port Qasim in Karachi. Karachi also hosts a solar-powered sovereign GPU facility through the Telenor and Data Vault partnership, indicating that energy-coupled compute is operationally feasible today; it can run AI workloads at a fraction of foreign cloud costs while keeping sensitive data within Pakistani jurisdiction. The next step is converting this proof of concept into a replicable, scalable mechanism.

The regulatory constraints can be addressed. Independent power producers are bound by agreements channelling generation into the national grid, but the Competitive Trading Bilateral Contracts Market (CTBCM) already permits generators to sell directly to consumers. Operationalising it for collocated compute would push surplus power straight to data centres, sidestepping politically fraught IPP renegotiation, with no contracts being reopened and only existing surplus being redirected. Transmission modernisation matters, but it is a separate, slower process; treating it as a prerequisite delays the immediate opportunity. Deploy energy-compute zones now, upgrade the grid later, and the capital these zones unlock can itself fund and incentivise modernisation.

Policies must connect assets to a specific goal.

Capacity payments are treated as the cost of energy mismanagement — Rs2.1 trillion a year, roughly $7.5 billion, paid to plants that often generate nothing. Viewed differently, they are a prepaid infrastructure asset: 10 to 12GW of surplus generation, financed and available, awaiting demand Pakistan can’t supply. AI data centres are among the few industries with power needs large enough and requirements high enough to absorb that supply where it sits. Some 100GW of new data centre capacity is expected globally by 2030, backed by up to $3tr. Pakistan needs a fraction. Securing 1-2GW, under two per cent of the global pipeline, would generate $700 million to $1.5bn a year in electricity sales, offset up to 20pc of the capacity payment burden and attract $10–25bn, several times the $1.64bn Pakistan recorded in net FDI in FY26. The surplus is a sunk cost either way. The choice is whether it remains a liability passed on to consumers or becomes an asset earning foreign exch­­-ange.

Pakistan has engineers, a growing tech sector, and power it already pays for. What is missing is policy alignment connecting the assets to a specific goal. Three interventions follow: 1) immediate: operationalise the CTBCM for collocated compute, notifying rules that let generators contract directly with off-grid data centres. This needs no new money and no IPP renegotiation, only a regulatory decision and clear direction-setting; 2) short-term: designate energy-compute zones at two or three sites where surplus generation sits stranded, with land, licensing and battery storage pre-cleared, so that projects like Gul Ahmed’s become a template to emulate rather than one-off interventions; 3) long-term: knit the sovereign compute capacity taking shape across Islamabad, Karachi and Lahore into a single national network, and link that GPU capacity to universities and startups so that Pakistan builds infrastructure and the talent to run it.

The generation is already there, as are the resources and the know-how. What remains is the will to act and an environment that turns it into tangible gains for everyone.

The writers are Fulbright scholars.

Published in Dawn, August 21st, 2026