How the University of Utah built a sovereign AI factory to accelerate breakthroughs and slash cloud costs

The scaling of artificial intelligence has forced IT leaders to re-evaluate infrastructure. While public clouds offer rapid deployment for general applications, they introduce steep trade-offs for highly regulated, data-intensive workloads. Issues like high latency, unpredictable operational costs, and diminished data control complicate the development of proprietary intellectual property. To bypass these limitations, leading institutions are pioneering a new approach: the sovereign AI factory.

At the University of Utah, leadership confronted this issue directly. The institution needed to boost its computational capacity to accelerate clinical and academic work without compromising safety. Because these research avenues rely heavily on sensitive patient records, genomic profiles, and highly regulated healthcare data, a public cloud architecture was insufficient. The university required complete data control, strict compliance, and high performance.

By collaborating with HPE and NVIDIA, the University of Utah designed and deployed an integrated, full-stack sovereign AI factory. This public-private-philanthropic co-investment—championed by the university, the State of Utah, and the Huntsman Family Foundation—serves as an example for CIOs managing high-stakes data environments.

The challenge: Balancing computational scale with data sovereignty

For any organization handling protected information, public cloud environments introduce significant regulatory compliance risks. For the University of Utah’s researchers at the Huntsman Cancer Institute and the Huntsman Mental Health Institute, moving vast biological and behavioral datasets to external infrastructure was impractical.

Traditional IT architectures also struggle under the weight of generative and agentic AI workloads. These applications require tightly integrated clusters where high-performance compute, ultra-low-latency networking, and resilient storage operate as a single system. Fragmented infrastructure creates data bottlenecks, stalls GPU utilization, and increases operational costs.

“Infrastructure is the engine behind AI-enabled innovation,” explains Professor Manish Parashar, Chief Artificial Intelligence Officer at the University of Utah. The fundamental challenge was building an infrastructure robust enough to run large language models (LLMs) and multi-modal datasets while keeping that data strictly contained within a secure, legally compliant perimeter.

The solution: An integrated, full-stack sovereign architecture

To solve this problem, the University of Utah implemented an HPE AI Factory with NVIDIA. Rather than piecing together components from multiple suppliers, the university chose a fully integrated stack managed through HPE GreenLake, combining cloud-like flexibility with the strict control of on-premises infrastructure.

The underlying hardware foundation consists of HPE Cray XD670 servers equipped with high-performance NVIDIA Hopper GPUs. This provides the massive parallel processing power necessary to ingest and analyze multi-modal research data.

The system’s control and operational layer relies on software integration:

• NVIDIA AI Enterprise: Provides secure, end-to-end software frameworks to develop, optimize, and run advanced machine learning models.
• HPE Morpheus and HPE OpsRamp Software: Streamlines multi-tenant operations, simplifies hybrid cloud resources, and allows the university to manage a shared platform for diverse research groups.

To address the power, density, and cooling demands, the infrastructure is hosted at a high-density colocation facility managed by DataBank. This setup balances performance requirements with energy efficiency and rapid scalability.

The benefit: High-velocity discovery and cost predictability

Moving away from the public cloud model has delivered significant structural benefits. By establishing complete ownership over the AI pipeline, the university can repatriate critical research workloads, allowing it to cut operating expenses by up to two-thirds compared to commercial public cloud alternatives. This cost reduction fundamentally changes the economics of long-term AI programs.

Crucially, the sovereign design eliminates data security trade-offs. Because information remains within a governed environment, researchers can safely analyze protected health information (PHI) and genomic data.

The platform also acts as a unified regional resource. Through the Utah Research & AI Infrastructure for a Statewide Ecosystem (RAISE) initiative, the university extends access to other higher education institutions, state agencies, and regional tech startups. This approach democratizes elite computing access and transforms the state’s “Silicon Slopes” tech corridor into a highly competitive hub for workforce talent and business investment.

The results: Accelerating real-world clinical breakthroughs

The most compelling justification for a sovereign AI factory is its impact on operational outcomes. At the University of Utah, the system has compressed the time required to process trial-and-error research questions, transforming months of data analysis into hours.

In clinical oncology, researchers utilize the platform to analyze vast biological datasets, mapping out the genetic drivers of rare cancers to design highly personalized patient treatment protocols. Simultaneously, the Huntsman Mental Health Institute processes massive volumes of behavioral and clinical data to identify early warning signs of mental health conditions.

“Our goal is ensuring the state is awash in computing power by building a robust and scalable AI ecosystem,” statesTaylor Randall, President of the University of Utah. By pairing infrastructure with clear public accountability, the institution proves that data control does not require sacrificing technological velocity.

Strategic takeaways for enterprise IT leaders

The University of Utah’s deployment offers several clear lessons for corporate technology executives:

1. Sovereignty can reduce risk and cost: For core intellectual property and protected data, cloud repatriation to a managed sovereign AI factory stabilizes operating margins and simplifies regulatory compliance.
2. Full-stack integration avoids bottlenecks: Disconnected components limit GPU efficiency. Selecting co-engineered stacks, like the HPE Sovereign AI Factory delivered via HPE GreenLake, minimizes deployment complexity and optimizes system throughput.
3. Density demands specialized facilities: Next-generation processors require specialized power and cooling capacities. Partnering with high-density colocation facilities is essential for maximizing operational lifespan and system efficiency.

For enterprise leaders, building an AI factory is no longer just an infrastructure upgrade; it is a critical strategy to protect data assets, control long-term operational costs, and accelerate proprietary innovation. For more information, visit hpe.com/ai. 

**********

As AI becomes increasingly central to economic competitiveness, scientific advancement, and national priorities, organizations require infrastructure that balances performance with security and sovereign control. Together, HPE and NVIDIA co-engineer rack-scale AI systems that integrate AI computing, high-performance networking, and supercomputing expertise to support large-scale AI workloads. This provides enterprises, governments, and research institutions with a trusted foundation for sovereign AI initiatives while maintaining control over critical data, models, and operations.