Pakistan launched Sky47 Karakoram-01 on 24 July 2026 and is deploying NVIDIA GPUs across public institutions to build national ai infrastructure. Here is how it compares to regional peers.
On 24 July 2026, Prime Minister Shehbaz Sharif and Field Marshal Syed Asim Munir jointly inaugurated Sky47 Karakoram-01 in Islamabad, Pakistan’s largest AI-ready sovereign cloud facility. The facility operates at 8.5 megawatts, is built to Tier III and Tier IV standards, and is designed to scale beyond 50 megawatts with more than 3,000 rack capacity. A 5-megawatt node at Karachi’s Port Qasim is under construction and expected to be ready by September 2026, with a Lahore facility planned to follow.
A week earlier, the National Telecommunication Corporation had signed a three-year agreement to deploy enterprise-grade NVIDIA GPU infrastructure, including the H200, A100, and B200 series, across secure government-operated data centres. The project is implemented in partnership with Data Vault Pakistan and will host multiple open-weight large language models, including Llama 3.1 and GPT OSS 20B, among other open-weight systems, alongside text-to-speech and inference services. NTC has committed to 99.9 per cent service availability.
These two initiatives arrived within days of each other and together represent the most ambitious public-sector AI infrastructure programme Pakistan has ever announced. What follows is an honest assessment of what they mean, what they do not yet deliver, and how Pakistan’s position compares to the regional peers it is competing with.
What Sky47 Karakoram-01 Actually Is
Sky47 is a sovereign cloud and AI computing facility, not simply a data centre in the traditional sense. The distinction matters. A conventional data centre stores and processes data. A sovereign AI facility is designed to run AI inference and training workloads under local jurisdiction, making it available to government agencies, banks, hospitals, and enterprises that need to process sensitive data without routing it through international cloud providers.
The facility offers sovereign cloud infrastructure, high-performance AI computing, cybersecurity solutions, enterprise hosting, and managed technology services under one platform. For Pakistani banks that must comply with State Bank of Pakistan data residency requirements, or for healthcare institutions handling patient data, or for government agencies working with classified information, Sky47 provides something that AWS, Google Cloud, and Azure do not: compute power that stays on Pakistani soil under Pakistani law.
The Prime Minister called the inauguration a digital milestone and a step toward a forward-looking Pakistan. Field Marshal Asim Munir’s presence signals the national security dimension of the investment: AI infrastructure is now treated as strategic infrastructure in the same category as defence and energy.
What the NTC GPU Programme Delivers
The National Telecommunication Corporation’s three-year GPU deployment is the infrastructure that will make AI workloads possible inside government institutions that currently lack the compute to run them. The H200, A100, and B200 series NVIDIA hardware represents genuine enterprise-grade AI capability, not the entry-level GPU provisioning that earlier public-sector AI initiatives in emerging markets have sometimes involved.
The model stack hosted on this infrastructure is notable for its deliberate choice of open-weight systems. Llama 3.1, Qwen3 32B, and Kimi K2 are all open-weight models that can be deployed, fine-tuned, and operated within controlled environments without the ongoing API dependency that proprietary cloud-hosted models require. Pakistan’s government is building AI capability it owns rather than renting inference from foreign providers.
NTC Managing Director Maj Gen (R) Ali Farhan described locally hosted computing resources as critical for protecting Pakistan’s digital sovereignty and supporting future government AI applications. The framing aligns directly with Pakistan’s National Data Governance Policy 2026, which requires government agencies deploying AI to maintain explainability, human oversight, and data residency within Pakistani borders.
How Pakistan Compares to Regional Peers
The honest answer is that Pakistan is behind but moving faster than at any previous point.
Saudi Arabia’s SDAIA, the Saudi Data and Artificial Intelligence Authority, has been building AI infrastructure under Vision 2030 since 2019, with investments in AI cloud capacity, national AI data platforms, and the Humain sovereign AI company launched in 2025 with a mandate to build AI infrastructure at national scale. Saudi Arabia’s sovereign AI investment runs into the tens of billions of dollars. Pakistan’s current programme runs into the hundreds of millions.
The United Arab Emirates has deployed the Falcon family of open-weight large language models through the Technology Innovation Institute and has built G42 into one of the world’s most significant sovereign AI companies, with data centre capacity across the Gulf, Africa, and Southeast Asia. The UAE’s AI infrastructure investment is backed by Abu Dhabi sovereign wealth capital of a scale Pakistan does not have.
India’s IndiaAI mission has committed to building a national AI compute infrastructure, including a 10,000-GPU supercomputer, with government funding. India’s IT services sector, at roughly ten times Pakistan’s scale in revenue terms, provides a tax and investment base that allows larger commitments.
The comparison is not discouraging. It is calibrating. Pakistan is not trying to match UAE or Saudi investment in the next three years. It is building the sovereign infrastructure foundation that did not previously exist, without which the next tier of government AI deployment and private-sector leverage of public compute would be impossible. Sky47 and the NTC GPU programme are first-layer infrastructure. The returns compound over the facilities’ operational lives, not at inauguration.
What Comes Next in Building National AI Infrastructure
Three developments will determine whether this infrastructure delivers its stated purpose or becomes another underutilised government investment.
The first is utilisation. AI infrastructure without use cases generates no return. The government needs to identify which ministries, regulators, and agencies will migrate workloads onto NTC’s GPU infrastructure and what those workloads will be, within a defined timeline, not as an aspirational note at the end of a press release.
The second is private-sector access. Sky47 explicitly targets not just government institutions but banks, hospitals, and technology firms. The commercial pricing model, the service reliability record, and the data sovereignty guarantees that Sky47 can offer to private-sector clients will determine whether the facility becomes part of Pakistan’s enterprise technology stack or remains primarily a government asset.
The third is talent. The engineers capable of fine-tuning Llama 3.1 on government data, operating H200 GPU clusters, and building AI applications on top of MCP-ready infrastructure are scarce globally and scarcer in Pakistan. The government’s commitment to training one million people in AI within three years is the demand-side investment that has to keep pace with this supply-side build-out.
Pakistan has built the facility. The harder work is building the ecosystem around it that makes the facility matter.
