On Friday afternoon, Prime Minister Shehbaz Sharif stood in Rawalpindi and switched on Sky47's Karakoram-01, a campus its owners are calling Pakistan's first purpose built, AI ready data center. The ceremony had the trimmings you'd expect: promises of near unbreakable security, talk of a bolder and more forward leaning Pakistan, and a new term from the IT minister, who described what she called token intelligence as the most important economic unit of today. The facility runs on 8.5 megawatts, with a second, smaller site planned for Port Qasim by September. It was built by Sky47 Limited, a subsidiary of Mari Energies through Mari Technologies, and it now offers sovereign cloud hosting and high performance computing to whoever can pay for it.
As a piece of engineering, it's genuinely impressive. As a use of national attention and capital right now, I think it's badly timed, and the case against it isn't ideological. It's arithmetic.
Start with power. Pakistan's electricity sector is sitting on circular debt north of two trillion rupees, a hole built out of unrecovered bills, transmission losses, and capacity payments to power producers who get paid whether their electricity is used or not. Ordinary households have absorbed tariff hikes of more than 550 percent since 2007 trying to close that gap, and the debt keeps growing anyway. A hyperscale compute cluster is a serious, round the clock electrical load. Adding one to a grid that already can't collect what it's owed doesn't repair anything. It just adds a demanding new customer to a system running on unpaid IOUs.
Then there's water. Pakistan's per capita water availability has fallen from roughly 5,600 cubic meters in 1956 to under 800 today, a level that puts the country firmly inside internationally recognized scarcity. More than eighty percent of the population already goes through severe water stress for part of the year. Data centers, even efficient ones, draw water for cooling at scale, and none of the announcements around Karakoram-01 mention its footprint once it's running at full load. That detail tends to surface later, after a facility is already running and someone downstream starts asking questions.
There's a currency angle too, and it gets mentioned the least. Pakistan is still inside an IMF program, still rebuilding reserves that fell to roughly 3.5 billion dollars during the 2023 crisis before climbing back toward 16 billion under strict conditions. Almost nothing in a facility like this, the chips, the cooling hardware, the networking gear, is made domestically. It's imported and paid for in dollars. Calling a facility sovereign when its physical stack depends on imported silicon is a stretch, and every dollar spent on it is a dollar the reserve buffer doesn't have for something else.
None of this means local AI infrastructure is a bad idea on principle. Pakistan reportedly spends somewhere between 700 and 800 million dollars a year renting computing power from foreign cloud providers, money that leaves the country before a single local startup or researcher benefits from it. Bringing some of that home is a reasonable goal. The real question is whether this specific project, built under these specific constraints, actually moves the country closer to that goal, or whether it just gives officials something to cut a ribbon on.
The Math Nobody Put In The Press Release
There's also something a little absurd happening around the word "first." Sky47 calls Karakoram-01 the first purpose built AI ready campus in the country. A company called Data Vault made almost the identical claim about a facility in Karachi back in mid 2025. Khazana Cloud and the Air Force run NASTP have separately announced plans for what they're calling Pakistan's first hyperscale data center, built on Huawei hardware. A fourth project, the Quantum Global Data Centre in Karachi, broke ground this summer with a 230 million dollar first phase and a 600 million dollar target by 2030. On the exact same day Sky47 opened its doors, a different ministry signed an agreement for yet another "Pakistan's first" AI center, this one for industrial manufacturing. When four or five organizations race to plant the same flag in the same year, that's usually not a coordinated strategy. It's a sign that everyone wants the headline.
That scattershot pattern matters because data centers are capital intensive, not labor intensive. Once fully staffed, Karakoram-01 will likely employ a modest team relative to the size of the investment. Most of the real spending goes to imported hardware, construction contractors, and the people who structure the financing. The bigger promise, that a project like this transforms the job market for ordinary Pakistanis, tends to outrun what any single facility can deliver.
And sovereignty is a strange word to reach for when the country still has almost no domestic large language models, limited AI research funding, and a small pool of machine learning talent that mostly ends up working for foreign companies anyway. A data center without a domestic AI ecosystem to fill it is a warehouse with good branding. Owning the building isn't the same as owning the intelligence running inside it.
What Would Actually Move Pakistan's AI Ambitions Forward
Grid Reliability
Reduce load-shedding and collection losses before adding a hyperscale customer to a system that already can't pay its own bills.
Broadband Access
Expand affordable connectivity so ordinary users can actually reach the infrastructure being built in their name.
School Compute Labs
Put devices and connectivity in classrooms now, since that's the workforce any future AI industry will depend on.
Domestic Research
Back Urdu language models and local research groups so the country produces intelligence, not just space to rent it out.
Startup GPU Credits
Route compute directly to researchers and small startups instead of leasing capacity mainly to whoever can already afford it.
Resource Disclosure
Require public water and energy impact reports before the next facility breaks ground, not after residents start asking questions.
None of this means the people behind Karakoram-01 are acting in bad faith, or that the broader push toward local compute is doomed. If grid reliability improves, if the water footprint gets disclosed honestly, and if the facility ends up serving Pakistani startups at fair prices rather than mostly large corporate clients, this could still turn into something worthwhile. That's a real possibility, and it's worth watching how the Port Qasim and Quantum Global expansions play out before writing the whole push off.
But a ribbon cutting and a slogan are the easy part. The hard part, fixing a grid that loses money on every unit it sells, stretching a water supply thinner than Saudi Arabia's, and building an AI talent base deep enough to justify the word sovereign, is the work nobody photographed on Friday. Until that happens, I'd hold off on calling this a win.