Pakistan crossed from strategy to infrastructure in thirteen months

Disclosure: the author holds a personal, unpaid honorary advisory appointment connected to Pakistan's investment promotion work. ImpactLab has no institutional partnership with the Government of Pakistan, no Pakistani public body reviewed this issue before publication, and no one outside this publication had editorial input into it.

Most national AI strategies do not survive contact with a budget cycle. They are published, praised, and then sit in a drawer while the ministry that wrote them moves on to the next document. Issue 11 of this publication made that point about Pakistan, and made it about several European member states in the same breath.

Thirteen months on, the Pakistani version has done something the European ones mostly have not. It has produced operating assets.

Count what now exists that did not exist in July 2025. A National AI Policy approved at cabinet level. A national AI fund constituted from Ignite's research and development allocation. The country's largest data centre — 8.5 MW, built with ZTE — inaugurated and running. A private data centre with USD 230 million committed behind it. A compulsory artificial intelligence course in every undergraduate and postgraduate degree programme in the country. A domestic Chromebook assembly line. A founding seat at a new multilateral AI body. Technology exports at USD 4.6 billion, a record, up roughly 21 percent year on year. And, on 18 August, the first physical office of the world's largest search company, inaugurated at the Prime Minister's House.

That is a fast thirteen months by any comparative standard. It is faster than the Nordic equivalent, and I say that as someone who has spent this year documenting why Norway — a country with a far higher per-capita income and a functioning regulatory apparatus — still has no AI act in force and no maintained inventory of the AI systems its own state is running.

The interesting question about Pakistan is therefore no longer whether the ambition is serious. On the evidence it is. The question is whether the pieces are being assembled into something that compounds, or accumulated as separate announcements. That is a coordination problem, and coordination problems are the cheapest category of problem a state can have.

The evidence

Exports are at a record and accelerating. Technology and IT-enabled services exports reached USD 4.6 billion in FY2025-26, up from roughly USD 3.8 billion the previous year — approximately 21 percent growth. Total services exports were USD 10.04 billion, up about 19 percent. July 2026, the first month of the new fiscal year, ran at USD 417 million in technology exports, 18 percent ahead of July 2025 ^[1]. Freelancers contributed on the order of USD 1.1 billion of the annual figure, with the Pakistan Software Export Board and subsequent reporting citing figures between USD 1.1 billion and USD 1.16 billion depending on how freelance income is scoped ^[2].

Connectivity ranking improved sharply. Pakistan rose 27 places in the ITU ICT Development Index, from 56.4 to 67.7 ^[3].

Compute exists now, not in a plan. Sky47's Karakoram-01 was inaugurated on 24 July 2026 — 8.5 MW, built with ZTE, and described by the government as the country's largest purpose-built data centre ^[4]. Separately, the Quantum Global Data Centre in Karachi, backed by Gul Ahmed Energy Group with Huawei Pakistan as technology partner, has USD 230 million committed initially and a reported path to USD 600 million ^[5]. In July the National Telecommunication Corporation entered a three-year arrangement to place accelerated computing capacity at the disposal of government institutions, reducing reliance on public cloud ^[6].

The education pipeline has been changed structurally. On 19 February 2026 the Higher Education Commission mandated a compulsory three-credit-hour AI course in every undergraduate and postgraduate degree programme in the country, from the 2026 academic session ^[7]. Very few states have done this. It is the single most consequential AI literacy decision any government made in the first half of 2026, and it received almost no international attention.

Foreign presence deepened. Google inaugurated its first physical office in Pakistan on 18 August 2026 at the Prime Minister's House, with a delegation led by Wilson L. White, Vice-President for Government Affairs and Public Policy ^[8]. Google's stated contribution to date includes 350,000 Career Certificate scholarships, of which 150,000 were announced in July 2026, alongside developer training and platform access commitments ^[9]. These are the company's own figures, published without methodology, and no independent verification exists — but the direction is not in dispute. A Chromebook assembly line was inaugurated on 4 November 2025 by the National Radio and Telecommunication Corporation, at a reported capacity of roughly half a million units a year ^[10].

And Pakistan took a seat rather than waiting for an invitation. On 16 July 2026 it became a founding member of the World Artificial Intelligence Cooperation Organisation, signed in Shanghai by the Deputy Prime Minister and Foreign Minister ^[11].

The hedge is an asset, if it is held deliberately

The obvious observation about July and August is that Pakistan joined a Chinese-led AI governance body and inaugurated a major American technology company's first office thirty-three days apart.

The reflexive reading is that this is incoherent. I think that reading is wrong, and the more useful one is available.

Issue 05 set out the sovereign-AI trilemma: compute control, allied alignment, citizen affordability — pick two, the third gives. Almost every state in the world is currently being asked to resolve that trilemma by choosing a bloc, and almost every state that has chosen has paid for it in optionality. The Gulf states bought frontier access with binding assurance commitments to a third government. The Nordics accepted soft compute control in exchange for European alignment and affordability.

Pakistan is, at present, among the very few large developing economies holding a working relationship with both AI blocs simultaneously: Chinese vendors building parts of the physical layer, American platforms supplying skills and distribution, and a founding seat in a multilateral body positioned explicitly around Global South representation. That is not a contradiction. It is the position most middle powers say they want and cannot get.

But an optionality position is only an asset if somebody is managing it. Held deliberately — with a published view of which relationship binds when they conflict, and what Pakistan asks of each — it is genuine leverage, and it is among the strongest cards a country of 240 million people with a large English-competent diaspora currently holds. Held accidentally, as two dependencies acquired separately by different actors, it becomes a liability the first time an export licence or a standards vote forces the question.

The distinction between those two states costs nothing to establish. It costs a document.

What the Google office actually tells us, read correctly

Two things about the August inauguration are worth separating, because the coverage conflated them.

The first is that Google disclosed no investment figure, no local headcount, no data centre and no cloud region. That is a fact and it should temper the celebration. What opened is a policy and distribution office, and the honest description of Google's Pakistan contribution to date is training and platform access rather than capital.

The second is more interesting, and it is a genuine Pakistani policy success that nobody has claimed credit for.

In July 2025 Pakistan enacted a 5 percent Digital Presence Proceeds Tax on foreign digital vendors with significant digital presence and no physical presence in the country. Within weeks, the Federal Board of Revenue confirmed that Google fell outside the levy on the basis that it operates through a registered local branch ^[12]. Local establishment also opens a path to long-run income-tax relief inside a Special Technology Zone.

Read that sequence as a policymaker rather than a critic. Pakistan built an instrument that made physical presence materially cheaper than remote supply, and thirteen months later the largest possible test case established physical presence. The tax code did what tax codes are supposed to do.

The lesson is not that the office was bought. It is that Pakistan has demonstrated it can move a global platform's operating footprint with a well-designed fiscal instrument. That is a much more encouraging finding than "Google turned up and brought nothing."

What would compound this

Three things, in ascending order of cost. None requires new legislation.

Publish the division of labour between the AI institutions. Pakistan now has a National AI Policy with a mandated council, a National AI Fund, an Islamabad AI Declaration, the National AI Advancement Initiative, and Ignite running both the incubation centres and the Pakistan Startup Fund. Every one of those is defensible on its own. Together they need one page stating who owns what, because an investor or a development finance institution assessing Pakistan cannot currently tell which body to talk to — and that ambiguity is priced into every conversation. Independent analysts flagged the coordination risk around the sovereignty declaration in February ^[13]. It is the cheapest thing on this list and the highest return.

Commission an independent evaluation of the incubation centres. The government reports strong results from the National Incubation Centres — more than 2,250 startups since 2017 and Rs 36.3 billion in revenue on the most recent account. Its own project page reports materially different figures ^[14]. I am not able to say which is right, and that is exactly the problem: a verified track record is an asset that attracts capital, and an unverified one is a discount applied to everything built on top of it. Pakistan almost certainly has a better story here than it can currently prove. Proving it is worth more than announcing the next programme.

Point the fiscal instrument at compute and engineering. The Digital Presence Proceeds Tax and the Special Technology Zone regime together demonstrated that Pakistan can change a multinational's footprint with tax design. The obvious next application is a graduated incentive tied to what the country lacks: local engineering headcount, a cloud region, or capacity reserved for Pakistani public bodies, researchers and small businesses. Issue 04 proposed exactly this structure for Norwegian clean-energy AI zones — a public-interest carve-out in exchange for a concession. It transfers straightforwardly, and Pakistan has better evidence than Norway that its version of the lever works.

The objection worth taking seriously

The strongest objection to everything above is that institutional velocity is not the same as capability, and that a country can accumulate policies, declarations, funds, hubs and inaugurations at speed while the underlying constraint — reliable electricity, reliable connectivity, foreign exchange, a data protection regime — goes unaddressed. On this reading the thirteen-month build is announcement inflation, and the export growth is a freelancing and services-outsourcing story that would have happened without any of it.

That objection has real force, and the connectivity evidence supports part of it. Internet service in Mirpur, in Azad Jammu and Kashmir, was restored in late July after a shutdown reported at between forty-five and forty-eight days depending on the outlet, and Pakistan still has no comprehensive data protection law ^[15]. Civil-society analysts have also documented how many schools lack the basic computing facilities that a national AI literacy push assumes ^[16] — a hard fact to hold alongside a compulsory university AI course.

The answer is that the two are sequenced, not opposed. Institutional capacity is what allows a state to fix physical constraints in a durable way rather than a project-by-project way, and Pakistan has spent thirteen months acquiring institutional capacity. The risk is not that this was the wrong order. It is stopping here — treating the institutions as the achievement rather than as the machinery for the harder work.

What ImpactLab is doing

The Norway–Pakistan publication track is the answer in publication mode: analysis published annually, with a named diaspora and academic contributor list. As stated in Issue 11, there is no institutional partnership counterpart and we are not claiming one.

The Middle East ATLAS carries the Arabic-language and Urdu-language vector for AI literacy work under Pillar I, with a scoping note publishing in the fourth quarter of 2026.

Both sit alongside AI for Norway as the small-state and large-state halves of one question: whether AI literacy compounds on comparable timescales at radically different scales. On the evidence of the last thirteen months, the large state is moving faster on the literacy instrument than the small one — a Pakistani undergraduate now has a mandatory AI course, and a Norwegian public body still has no supervisory guidance. That is not the direction anyone predicted, and it is the most useful thing in this issue.