Building a Sovereign GenAI Stack for the United Nations with Intel and OPEA

In a world increasingly shaped by generative AI (GenAI), the United Nations (UN) has taken a bold step toward digital sovereignty by developing an open-source AI infrastructure in collaboration with Intel and the Open Platform for Enterprise AI (OPEA).

The goal? To deploy a scalable, vendor-neutral GenAI solution tailored to the unique needs of global institutions — one that aligns with the UN’s mission of independence, data security, and sustainability.

The Challenge

At the heart of this initiative is the AI Sovereign Stack, a cloud-agnostic and hardware-flexible AI framework designed to run across on-premises and cloud environments. It leverages open-source technologies and Intel’s hardware-optimized software to support a wide range of AI workloads, including embedding, text generation, and retrieval-augmented generation (RAG).

One of the key challenges the UN faced was building an AI solution that maximizes resource efficiency while minimizing dependency on costly, power-intensive hardware. While GPUs offer strong performance, they’re not always practical for every use case. Intel’s contribution — particularly through Intel® Xeon® CPUs with Intel® Advanced Matrix Extensions (Intel® AMX) — made CPU-based inference a viable and cost-effective alternative for many GenAI tasks.

The Solution

To ensure maximum interoperability and flexibility, the architecture integrates seamlessly with the OpenAI Chat Completions API, allowing UN teams to reuse existing tools and workflows with minimal changes. Core components like vLLM and Text Embedding Inference (TEI) enable scalable serving of large language models (LLMs) and embeddings across CPU, GPU, and AI accelerators. These microservices are modular and containerized, allowing the UN to adapt to new models or use cases quickly.

The deployment, led by Intel and guided by OPEA’s microservices blueprint, was completed in the UN’s data center. LLaMA 1B and 8B models were deployed across CPU and GPU environments, respectively, demonstrating the stack’s flexibility. The front-end, powered by Open WebUI, offered secure, self-hosted chatbot interfaces enriched with UN-curated datasets and domain-specific system prompts.

As the system matured, the UN team expanded its model portfolio to include open-source models like Mistral, Qwen, and DeepSeek, showcasing the architecture’s extensibility. Secure access was ensured via VPN and enterprise directory integration, keeping all inference and user data within a tightly controlled environment.

Conclusion

The success of the AI Sovereign Stack highlights a growing need for responsible, open-source AI infrastructure that respects data privacy and avoids vendor lock-in. As organizations across the globe seek to embrace GenAI, the UN’s approach — supported by Intel and OPEA — provides a compelling blueprint for sovereign, future-ready AI deployments.

You’re invited to read more about this exciting collaboration, including architecture and deployment details in the full white paper.

You’re also invited to review the solution on the UN Site: Unite Ideas - Idea: ‪United Nations x Intel x OPEA: Text Generation and Embedding‬.

Notices and Disclaimers

Performance varies by use, configuration, and other factors. Learn more on the Performance Index site.
Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available ​updates. See backup for configuration details. No product or component can be absolutely secure.
Your costs and results may vary.
Intel technologies may require enabled hardware, software, or service activation.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.