Exporting surplus power as AI compute

Surplus availability of power is a good problem. Capacity payments that amount to almost Rs2 trillion continue to take a toll on the economy, but there exists an opportunity to turn this costly surplus into a strategic export. The consumer pays for megawatts that we don’t use, but there is a shortage affordable power globally to run artificial intelligence (AI) training models, data centres, and more.

An ambitious endeavor no doubt, but we have the core ingredient to utilise surplus power and establish a firm power intensive digital infrastructure, eventually resulting in lower costs for the consumer, while cementing foundation of a more forward-looking export-oriented industry; a far cry from our current industry which belongs in the 18th century.

There exist a case for the establishment of AI parks that can host large digital load (LDL) campuses, co-located with existing power plants that have low dispatch rates, and linked with multiple submarine cables. This could be an accidental renaissance for Pakistan, where availability of surplus power in the country directly coincides with a shortage elsewhere in the world.

Demand for compute is exploding globally and utilities in the US and Europe are scampering for power availability of firm power — power-producing capacity that is available at all times even in adverse conditions — that doesn’t exist, or continues to increase in cost. Timelines are further stretched by interconnection queues, lead times for development of requisite infratructure, and fuel constraints. This leads to a scenario where price spreads are favourable in Pakistan where locations in the US have retail prices in the range of 12 to 18 cents per kWh, while dedicated data centres struggle to get availabiltiy of firm power at less than 8 cents.

Where the US has retail prices in the range of 12 to 18 cents per kWh, the marginal price of power plants located at or near the coast in Pakistan is around 6 cents per kWh

Pakistan already has the power infrastructure in place, with most of the surplus infarstructure along the coast, ideally located in close proximity to water, and connectivity through submarine cables. The marginal price of power plants located at or near the coast is around 6 cents per kWh.

There is an excess of over 4,000 MW of surplus power in the southern part of the country, which will only continue to increase as solarisation continues to strip grid demand. The exodus from the grid may further accelerate as battery prices continue to decline. The power-load curve has already started mimicking a duck curve as the load from the grid significantly drops during the hours when the sun is shining.

The grid needs firm demand, and it needs it fast; any delays on this front may lead to a grid death spiral, recovery from which would entail significant fiscal costs, which inadvertently would be borne by taxpayers. The days when price can be determined through a linear cost-plus structure, belong to the last century. Sticking to the same for the future would be a fallacy that only accelerates the death spiral.

More importantly, an AI park is not a real-estate play, and care must be taken to ensure it does not morph into one. It is a power and fibre play that ramps up firm and dedicated demand, easing out capacity payments, while opening up an export avenue to generate precious foreign currency inflows.

A review of publicly available data to rank potential sites on the basis of backhaul redundancy to connect to the world, interconnection strength, energy price and fuel security, and cooling/land practicality positions the Karachi-Hub-Jamshoro arc as ideally positioned for development of a flagship AI park, that spearheads the story of “exporting compute”.

There is more than 3,000 MW of power capacity that is located within reach of multiple cable landing stations, as well as access to water for cooling. Such a location also avoids the north-south transmission constraints, while providing access to power at a globally competitive price.

This also opens up potential to increase solar and battery penetration through policy interventions, thereby freeing up grid capacity to be reallocated to dedicated export-oriented power load. The threat to the grid posed by decentralised power can be turned into an advantage through tactical policy interventions that can encourage decentralised solar in the morning, and virtual power plants through batteries in evening. Instead of a power planning approach that belongs to the last century, a more portfolio-oriented approach is required that looks into the future.

The core hard elements for executing a successful compute export strategy are in place, further augmented by an effective foreign policy. The soft, policy elements is where the trouble lies. The policy interventions are focused, and finite, but the ability of regulatory bodies to execute the strategy, instead of creating yet another white elephant, leaves a lot to be desired. It requires notifying a new LDL class of consumer, finalising use-of-system charges in a way that reflects the economics of the underlying transaction, and more importantly, positioning the Competitive Trading Bilateral Contracts Market in a way that it looks like an actual competitive market.

It also requires a serious review of Personal Data Protection laws, in the absence of which we can never have a robust, and export-oriented compute, or data strategy. Similarly, it is also critical that a clear distinction is created that an AI park is not a telecom service, but rather a critical information infrastructure, and needs to be treated accordingly. If attracting foreign direct investment is actually a priority, then ensuring that equipment can move without any duties, or dividends can be repatriated without waiting in a queue for quarters at a stretch should be base-level considerations.

Speed matters in this context; any delays, or passing the hot potato will only penalise the taxpayer to foot the bill for yet another policy failure.

The writer is an assistant professor of practice at IBA, member of the Thar Coal Energy Board, and CEO of NCGCL

Published in Brackly News, The Business and Finance Weekly, September 29th, 2025