MAYS IBRAHIM (ABU DHABI)
The Mohamed Bin Zayed University of Artificial Intelligence (MBZUAI) will release the UAE’s first fully sovereign, domestically developed large language model K2-V2 on Tuesday, according to Richard Morton, Managing Director of Institute of Foundation Models (IFM) at the MBZUAI.
The move comes as part of the UAE’s push to establish an “AI factory” infrastructure, aimed at localising the full AI value chain — from data and compute to model development and applications.
K2-V2 will serve as the UAE’s sovereign base model “which is going to train all other models”, Morton told the Machines Can Think Summit 2026 on Monday.
K2-V2 is a 70-billion-parameter model in IFM’s K2 large language model series and is the MBZUAI’s most capable fully open model to date.
It follows what the institute calls a “360-open” approach, publicising weights, data composition, training methods, checkpoints, and fine-tuning recipes.
According to the MBZUAI, the model rivals leading open-weight systems in its class, outperforming Qwen2.5-72B and approaching the performance of Qwen3-235B, while maintaining full openness and reproducibility.
It is designed as a reasoning-centric foundation, with strong capabilities in long-context consistency, mathematical reasoning, and complex task execution, positioning it as a base layer for more specialised systems.
Crucially, K2-V2 will now serve as the sovereign base model for downstream UAE systems, including K2Think — the MBZUAI’s math and reasoning model, which had previously been trained on a foreign base model.
“We’ve retrained it completely on the UAE model, and it’s now 100% sovereign,” Morton told Aletihad on Monday. “Now the UAE has a sovereign model from the base all the way to its finely tuned math and reasoning layer. That’s a fundamental part of the UAE’s AI infrastructure.”
Morton has framed this shift as part of a broader transition toward systems that actively generate intelligence rather than simply host compute.
He argued that the UAE is “the perfect place” to build a national AI factory due to its access to land, energy, water, capital and “strategic leadership that is centralised, focused, reliable, and trusted”.
The IFM itself is an AI factory, producing the sovereign models that will underpin future AI systems across government, industry and society, he explained.
In an interview with Aletihad on the sidelines of the summit, Andrew Jackson, Group Chief AI Officer at G42, explained that this emerging system involves a two-layer structure: a “token factory” and an “agent factory”.
“The token factory is the data centre layer models producing tokens at scale,” he said. “The agent factory is building AI agents that solve real-world problems inside organisations.”
At the core of the token factory is Stargate UAE, a 1-gigawatt AI data centre cluster under construction in Abu Dhabi.
It forms part of the wider UAE–US AI Campus, a 5-gigawatt, 10-square-mile infrastructure zone being developed with partners including OpenAI, Oracle, Nvidia, SoftBank, and Cisco.
The first 200MW phase is scheduled to come online in 2026, powered by a mix of nuclear, solar, and natural gas energy.
The campus will provide nation-scale compute, low-latency inferencing and large-scale AI training capacity.
Last week, G42 CEO Peng Xiao told Bloomberg that the UAE expects its first shipments of advanced AI semiconductors within months, enabling the first Stargate phase to go live, with further expansion planned at up to 500MW per quarter as part of a long-term build-out to 5GW.
Beyond national sovereignty, Morton believes the UAE is positioning itself as a regional and global AI service hub, particularly for countries without the resources to build large-scale AI infrastructure.
The MBZUAI has already developed national language models for other countries, including Nanda for India and SHERKALA for Kazakhstan, offered as open resources.