Core42 announces new AI & cloud capabilities to accelerate enterprise transformation View announcement →
Core42 — A G42 Company
Case study · AI Cloud

Engineering the future of AI research at scale

How MBZUAI, the world's first graduate research university dedicated to AI, trains frontier models on Core42 AI Cloud.
MBZUAI campus
Watch the story · 3 min
Inside MBZUAI: training frontier models on sovereign infrastructure.

“Infrastructure bottlenecks have disappeared. When researchers need compute, it is available. Training cycles shorten and publication timelines accelerate.”

MBZUAI × Core42 AI Cloud
Researchers at MBZUAI

About MBZUAI

MBZUAI is a research-focused university in Abu Dhabi, and the first university dedicated entirely to the advancement of science through AI. The university empowers the next generation of AI leaders, driving innovation and impactful applications of AI through world-class education and interdisciplinary research. In 2025, MBZUAI launched its first ever undergraduate program, a Bachelor of Science in AI, with two distinct streams: Business and Engineering.

The challenge

Frontier research places unique pressure on infrastructure.

Training frontier models like Jais and K2 requires massive computational scale, with thousands of GPUs working through distributed training frameworks on high-performance, low-latency networks, where milliseconds of delay compound across runs spanning days or weeks.

Scale alone was not sufficient. Different research workloads have different optimal compute profiles. Frontier model training demands NVIDIA clusters, large exploratory workloads run efficiently on AMD infrastructure, and high-throughput inference suits specialized architectures such as Cerebras. Research directions shift quickly, and workloads had to move across accelerators without reengineering.

Sovereignty added critical complexity. Models designed for national use cases, such as Arabic LLMs and climate intelligence, must be trained and stored to meet strict regulatory and security expectations. And the infrastructure had to keep up with academic timelines, with consistency across the entire pipeline from first experiment through full-scale training and production inference.

Breakthrough ideas need immediate access to performance, scale, and sovereignty.
The approach

AI Cloud, the engine of discovery.

Core42 AI Cloud provided MBZUAI with a sovereign, heterogeneous compute fabric built for large-scale scientific workloads. The platform combines multiple accelerators and unified operations to support rapid experimentation and training in a single environment.

Multi-accelerator infrastructure

NVIDIA GPUs configured as a DGX SuperPod, hosted entirely within UAE jurisdiction, alongside AMD capacity providing massive cost-optimized compute.

Unified platform layer

All workloads are submitted through a single control plane that abstracts the complexity of a multi-accelerator environment. Scheduling, routing, and resource allocation are handled automatically.

High-performance foundations

A network fabric optimized for gradient synchronization keeps thousands of GPUs coordinating efficiently. Storage uses memory-mapping, asynchronous writes, and structured formats that prevent I/O bottlenecks, reducing time-to-train as workloads scale.
The results

Research velocity, unleashed.

MBZUAI has trained and deployed multiple frontier models, each representing months of intensive computation compressed into timeframes that match the pace of AI research.

Jais

The world's highest-quality Arabic large language model, serving 274 million Arabic speakers.

Jais Climate

The world's first bilingual LLM dedicated to climate intelligence, fine-tuned on 1.4 million climate instructions and launched for COP28 UAE.

K2

Matches frontier model performance at a fraction of the size, delivering faster inference at lower cost.

Specialized portfolio

Nile-Chat, Atlas-Chat, SHERKALA, NANDA, Vicuna, FM for Bio (GET), and World Model, each optimized on different accelerators.

Published research demonstrating computational capability rivaling top global institutions attracts the faculty and students who drive the next breakthroughs. Prospective faculty see infrastructure that enables rather than limits ambition, and students train on the same systems used by leading AI laboratories worldwide.

Each workload runs on the accelerator with the best performance and cost profile. AMD supports large exploratory research while NVIDIA clusters are reserved for frontier-scale training. At deployment scale, those efficiencies fund additional research projects, and complete sovereign control keeps research breakthroughs UAE and MBZUAI assets.

Takeaways

Sovereign scale delivers global-grade performance

Frontier-level performance, customer managed data zones, and compliance with UAE cloud policy can coexist when the cloud is designed for sovereignty.

Multi-accelerator design is strategic flexibility

Shifting workloads across AMD and NVIDIA gives researchers flexibility, efficiency, and future-proofing through one unified operating layer.

Unified operations are now essential

Consistent performance from early experimentation to production inference shortens the path from idea to deployment.

A platform that scales with research

The compute fabric, accelerator choice, and operational layer that global labs need to train large models and publish at competitive pace.

A roadmap built for next-generation hardware

Worldwide capacity growth and continuous upgrades to emerging accelerators support multi-year research programs.

Scale your research on sovereign AI infrastructure.

Request demo