Oracle founder and chief technology officer Larry Ellison has outlined what he sees as the biggest limitation facing today’s most advanced artificial intelligence models. According to Ellison, systems developed by companies such as Google, OpenAI and Meta are all built on essentially the same foundation. They are trained on publicly available internet data, which makes them increasingly similar and harder to differentiate.
Ellison made these comments during Oracle’s fiscal Q2 2026 earnings call in December. While acknowledging that training large models on open web data has already created what he described as the largest and fastest-growing business in human history, he argued that this phase is only the beginning. In his view, AI is now entering a second and potentially far more valuable stage.
That next phase, Ellison said, will be defined by AI systems that can reason over privately owned enterprise data without compromising security or confidentiality. Public data may be useful for building general intelligence, but it does little to help companies solve highly specific business problems. True value, according to Ellison, lies in giving AI controlled access to sensitive corporate information such as financial records, healthcare data, supply chain systems and internal operations.
Ellison believes this shift will dwarf the current boom around GPUs and data centres. He has argued that once models can securely interact with private data in real time, their usefulness and economic impact will increase dramatically. In his words, the opportunity created by this second phase will be even larger and more valuable than the infrastructure build-out that dominates the AI conversation today.
Oracle is positioning itself as a central player in this transition. The company’s core argument is that much of the world’s most valuable private data already sits inside Oracle databases. To capitalise on that, Oracle has been developing its AI Data Platform, which allows major AI models to query enterprise data securely using techniques such as retrieval-augmented generation. This approach enables AI systems to generate answers based on live private data without exposing or copying that information into the model itself.
To support this strategy, Oracle is dramatically increasing its spending. The company now expects to invest roughly $50 billion in capital expenditures for the full year, up from an estimated $35 billion just a few months earlier. At Oracle AI World in October, the company announced several large-scale infrastructure projects, including a 50,000-GPU AI supercluster powered by AMD MI450 chips, scheduled to launch in the third quarter of 2026. Oracle also revealed plans for its OCI Zettascale10 supercomputer, which will connect hundreds of thousands of NVIDIA GPUs.
Despite Oracle’s confidence, Ellison’s thesis is not without competition. Some experts believe synthetic data generation could reduce reliance on proprietary enterprise datasets. Others argue that real-time user interaction data from consumer platforms may ultimately prove more valuable than static corporate records. Meanwhile, cloud rivals such as Amazon Web Services, Microsoft Azure and Google Cloud are all racing to build their own enterprise AI stacks.
Oracle counters that its long-standing dominance in enterprise databases gives it a structural advantage that others lack. By late 2025, the company said its cloud backlog had crossed $500 billion, with AI-related demand accounting for the majority of that figure. If Ellison’s vision proves correct, the battle over AI may shift away from who has the biggest model, and towards who controls secure access to the world’s most valuable private data.
Discover the latest Business News, Sensex, and Nifty updates. Obtain Personal Finance insights, tax queries, and expert opinions on Moneycontrol or download the Moneycontrol App to stay updated!
