Pakistan is training Buraq, an Urdu language AI model, and wants the data kept at home
IT minister Shaza Fatima Khawaja told a World Bank panel in Seoul that the government is building its own model on Urdu and regional languages and buying sovereign computing capacity to run it.
Pakistan is building Buraq, a large language model trained on Urdu and other regional languages, and buying sovereign computing capacity so that Pakistani data is processed inside the country. IT minister Shaza Fatima Khawaja announced both in Seoul at the Global AI and Digital Summit, which ran from 19 to 22 October and was hosted by the South Korean government and the World Bank Group.
Urdu is the reason. It makes up only 0.03 percent of the web content gathered by Common Crawl, TechJuice reported, which is why most large models handle it poorly. The government chose to build from scratch rather than rely on foreign systems. Buraq has no website, public demo, release date or technical specification yet. Khawaja spoke on a World Bank panel about delivering government services digitally, and the announcement comes soon after Pakistan launched its first local AI cloud through a partnership between Telenor Pakistan and Data Vault Pakistan, TechJuice reported.
Four foundations under the model
In a Ministry of IT statement, Khawaja described Buraq and the public computing capacity as the AI layer of a programme resting on four foundations. The first is identity. The National Database and Registration Authority holds verified identity for more than 100 million Pakistanis, and that is being extended into a digital identity that authenticates citizens online and records and enforces their consent.
Payments are the second. The State Bank's Raast instant payment system is meant to turn an AI model's identification of a qualifying household into an actual transfer. The third is data. WASL, the national data exchange layer, is being built under the World Bank supported Digital Economy Enhancement Project and lets institutions share data, with every transaction authenticated, encrypted and logged. A citizen Super App and a National Open Data Ecosystem sit on top. The fourth is the National Data Governance Policy, which sets one rulebook for classifying, sharing, protecting and publishing public data.
“AI is the last layer, not the first.”
Governments that skip these foundations, she said, end up with pilots that never scale and systems that citizens do not trust.
Money and targets
In her keynote, titled Digital and AI Compacts: From Ambition to Action, Khawaja set out a $1 billion funding plan for AI through 2030, covering AI development, regulation and public sector projects such as welfare, healthcare and tax. It carries two targets: digital training for one million professionals outside the technology sector, and 1,000 fully funded AI doctorates. Both are goals, not results so far.
