Enterprises are increasingly rejecting the idea that they must depend on a handful of big tech companies to power their artificial intelligence systems. Instead, organizations are turning to open-source AI models and building what's known as "sovereign AI" infrastructure, where they maintain direct control over their data, intellectual property, and how their AI systems behave. This shift reflects a fundamental change in how businesses think about AI deployment.

Why Are Companies Suddenly Focused on Sovereign AI?

For the past few years, enterprises have relied heavily on proprietary AI models from major cloud providers. But this approach comes with tradeoffs: companies lose visibility into how their data is used, face unpredictable costs as AI usage scales, and have limited ability to customize models for their specific needs. Deloitte's new Open Model Engineering practice, launched on September 2, 2026, addresses these pain points by helping organizations design and deploy AI systems built on open-source frameworks and open models.

The practice reflects a broader recognition that enterprises need flexibility. Organizations today are navigating four critical considerations when deploying AI: the ability to choose the right model and architecture for each workload, predictable cost economics as AI usage grows, sovereignty over where and how models are deployed, and control over enterprise data, intellectual property, and model behavior with transparency on how the AI reaches its conclusions.

How to Build a Sovereign AI System: Key Steps for Organizations

• Model and Architecture Selection: Organizations can choose from proprietary models, open models, AI agent platforms, cloud services, and on-premises infrastructure depending on their specific workload requirements and regulatory environment.
• Cost Management Through Token Economics: By selecting the right model, infrastructure, and deployment patterns, enterprises can improve return on investment and avoid surprise costs as AI usage scales across their organization.
• Customization for Local Context: Open models can be fine-tuned for geographical, linguistic, and cultural context, allowing companies to build AI applications tailored to specific regions or business units.
• Data Protection and IP Control: Organizations gain greater choice over where models run, what data is uploaded to external systems, and how competitive advantages are retained by protecting intellectual property and the uniqueness of AI-generated outputs.

Deloitte's Open Model Engineering practice will initially focus on enterprise AI applications built on NVIDIA Nemotron open models and NIM microservices, which are specialized software components that make AI models easier to deploy and manage. The firm plans to hire, train, and certify forward-deployed engineers globally through fiscal year 2027, with these specialists working directly alongside clients to implement open model solutions.

What Does Sovereign AI Infrastructure Actually Look Like in Practice?

Real-world examples demonstrate that sovereign AI systems can be deployed quickly and operated reliably at scale. Penguin Solutions, an infrastructure specialist, designed and deployed the Haein cluster in collaboration with SK Telecom, one of Korea's largest sovereign AI clusters. The Haein cluster brings together 1,032 NVIDIA B200 graphics processing units (GPUs), 10 petabytes of high-performance storage, and 50 miles of networking to support demanding AI workloads across Korea's AI ecosystem while reinforcing data sovereignty and in-country operational control.

The deployment timeline illustrates the feasibility of building sovereign AI at scale. The Haein cluster reached production status in 60 days, launched with 100 percent of nodes within the target performance range, and after one year in operation has maintained greater than 99 percent hardware availability and 100 percent service level agreement compliance. This real-world success story shows that organizations don't need to choose between speed and control when building sovereign AI infrastructure.

"Organizations need flexibility in how they build and use AI. Deloitte's Open Model Engineering practice helps clients choose the right mix of AI models, cloud services, and on-premises infrastructure for each workload," said Sundhar Sekhar, Chief Services Officer at Deloitte.

Sundhar Sekhar, Chief Services Officer, Deloitte

Penguin Solutions offers a full-stack approach to sovereign AI deployment through its OriginAI infrastructure solution, which delivers validated AI factory reference designs for scalable AI training and inference workloads. By integrating compute, memory, storage, networking, security, and intelligent cluster management software, OriginAI provides a production-ready foundation to help organizations accelerate deployment while supporting compliance, performance, and long-term growth.

Who Benefits Most From Sovereign AI Approaches?

Different organizations have different reasons for pursuing sovereign AI. State-sponsored initiatives can establish sovereign AI environments with the control, policy compliance, and performance needed to support national research and public-sector operations. Neoclouds, which are regional cloud providers, can offer sovereign AI environments and capacity as a service with validated architectures and rapid deployment designed to support secure, multi-tenant AI workloads.

For enterprises, the appeal is straightforward: maintaining control over critical data, protecting intellectual property, and reducing third-party risk while supporting deployed AI initiatives at scale. As organizations evaluate their AI strategies, the shift toward open models and sovereign infrastructure represents a fundamental recognition that one-size-fits-all proprietary solutions no longer meet enterprise needs.

"Enterprises need both open and proprietary models for different workloads. Deloitte's Open Model Engineering practice, built on NVIDIA Nemotron models, gives developers greater control over enterprise data and proprietary information while enabling secure, responsible deployment of AI agents," explained Kari Briski, Vice President of Generative AI at NVIDIA.

Kari Briski, Vice President of Generative AI, NVIDIA

The momentum behind sovereign AI reflects a maturing market where enterprises are moving beyond early-stage AI experimentation toward production systems that must align with regulatory requirements, cost constraints, and strategic business objectives. As Deloitte expands its Open Model Engineering practice across North America, Europe, and Asia Pacific, and as proven deployments like Haein demonstrate operational success, the path toward enterprise AI independence becomes increasingly clear.

" }