EPRI: Local AI for Energy Research

Why It Matters

Local AI gives energy researchers secure, high‑performance compute without exposing sensitive data, setting a model for regulated industries seeking data sovereignty and faster insight generation.

Key Takeaways

• Dell Pro Max GB10 runs up to 200 B‑parameter AI models locally
• 128 GB unified memory enables cache‑augmented inference for Power Chat
• On‑prem AI cuts cloud cost uncertainty and data‑movement restrictions
• System delivers 1 petaFLOP performance, 35 tokens/second single‑user
• EPRI’s Power Chat provides instant document‑grounded answers within seconds

Pulse Analysis

The shift toward on‑premise AI in the energy sector reflects a broader industry need to reconcile powerful machine‑learning capabilities with stringent data‑security mandates. EPRI’s adoption of Dell’s Pro Max workstation, powered by NVIDIA’s Grace Blackwell GB10, demonstrates how a single rack‑scale system can host models rivaling cloud‑based supercomputers while keeping proprietary grid data behind corporate firewalls. This architecture mitigates the unpredictable expense of cloud compute, eliminates latency caused by data transfer, and satisfies regulatory oversight that often bars sensitive operational information from leaving secure facilities.

At the heart of EPRI’s deployment is Power Chat, a document‑grounded AI assistant built on the open‑weight gpt‑oss‑120b model. Leveraging 128 GB of unified memory and NVIDIA’s vLLM inference engine, the system caches encoded documents, enabling near‑instant retrieval and response generation. In practice, a single user experiences 35 tokens per second, and the platform scales to multiple concurrent users with modest performance degradation, proving that high‑throughput, low‑latency AI is achievable without massive data‑center footprints. This capability accelerates research cycles, allowing engineers to query dense technical manuals and simulation results in real time.

The implications extend beyond energy research. Industries such as finance, healthcare, and defense grapple with similar constraints around data sovereignty and compliance. By showcasing a compact, high‑performance AI solution that resides on‑premises, EPRI provides a blueprint for organizations to harness advanced analytics while maintaining full control over their data assets. As regulatory pressures intensify, the combination of Dell’s hardware and NVIDIA’s AI stack is poised to become a cornerstone for secure, scalable AI deployments across regulated sectors.