Steel, silicon, and Al: Resetting digital infrastructure

Oracle is engineering a new class of digital foundations by unifying dense GPU compute, next-gen cooling, and fast network fabrics to support Al at scale.

- VOICE&DATA BUREAU

Artificial Intelligence will not replace humans—it will amplify their ability to solve complex, global challenges. This was the central theme of Oracle Chairman and CTO Larry Ellison's keynote at Oracle Al World 2025, held in Las Vegas. Framing Al as the next Industrial Revolution, Ellison argued that Al's most important breakthroughs will happen not in internet search or chatbots, but in healthcare, climate, agriculture, and financial systems.

“The real opportunity is not just building these extraordinary electronic brains, but using them to solve humanity's most difficult problems,” Ellison said. He emphasised that the rise of Al necessitates more than just algorithms. It demands a robust and resilient digital infrastructure, capable of supporting a new kind of reasoning system.

To meet these demands, Oracle is expanding its Oracle Cloud Infrastructure (OCI) footprint, including a new AI supercluster in Abilene, Texas. This cluster will ultimately house over 450,000 NVIDIA GPUs and integrate its own power plants, high-speed fibre networks, and industrial-scale liquid cooling systems. “We are building billion-watt power plants and connecting them directly to our data centres,” Ellison said, calling it a new kind of infrastructure project—on par with railways and national grids.

Oracle's approach aims to unify AI infrastructure with real-world enterprise applications, enabling customers to securely run generative and reasoning AI models, including ChatGPT, Grok, Gemini, and Llama, alongside their private datasets.

PRIVATE DATA, NOT JUST PUBLIC MODELS

A major focus across sessions at Oracle AI World was the shift from public, general-purpose AI to enterprise-specific reasoning systems. While current models rely heavily on public internet data, Oracle is betting on Retrieval-Augmented Generation (RAG) and vector databases to allow customers to combine public knowledge with proprietary business data securely.