Oracle founder Larry Ellison Warns Big Tech AI Models Are Hitting a Wall

Dwijesh t

In 2025 and early 2026, Oracle founder Larry Ellison has emerged as one of the most vocal critics of the current direction taken by artificial intelligence models developed by Big Tech giants such as Google’s Gemini, OpenAI’s ChatGPT, and Meta’s Llama. While acknowledging their technical brilliance, Ellison argues that these models face a fundamental limitation: they are all trained on the same finite source of information the public internet.

The “Commoditized Internet Data” Problem

According to Ellison, the widespread reliance on publicly available web data has created the fastest-growing technology sector in history, but it has also capped the real-world business value of modern AI. Because companies scrape largely the same websites, books, and code repositories, their models increasingly suffer from what Ellison calls “zero differentiation.” In effect, most large language models know the same things, share the same blind spots, and struggle to offer a unique competitive advantage.

Ellison also highlights a major “context gap.” While models like GPT-4 or Gemini 2.5 excel at general knowledge, they lack awareness of how individual businesses actually operate. As Ellison has bluntly stated, ChatGPT does not understand a company’s internal accounts, customer orders, or supplier contracts precisely the data that matters most to enterprises.

Compounding the issue is a security paradox. Businesses want the reasoning power of advanced AI models but cannot risk uploading sensitive financial, operational, or customer data into public AI systems where it could be exposed or reused for training.

Oracle’s Vision for the Second Phase of AI

Ellison believes AI is now transitioning from its “training” phase into a far more valuable “reasoning” phase. In this next stage, the goal is not to build smarter general models, but to apply existing models to private enterprise data in secure environments.

Instead of sending data to AI, Oracle is focused on bringing AI to the data. The company is heavily investing in Retrieval-Augmented Generation (RAG), which allows models such as GPT-5 or Llama 4 to query and reason over private databases without the data ever leaving a company’s secure infrastructure.

Ellison is also championing the concept of “sovereign AI,” where governments maintain centralized, secure data repositories covering everything from healthcare to infrastructure that AI systems can analyze to solve national-scale problems like early cancer detection or food security.

Infrastructure as the Ultimate Moat

Finally, Ellison emphasizes that AI’s future depends on physical scale. He recently highlighted Oracle’s development of a massive 1.2-billion-watt AI data center campus in Texas, designed to serve as an “AI brain” capable of reasoning across the world’s most sensitive data.

In Ellison’s view, the winners of the next AI era will not be those with the biggest models, but those with the secure data, infrastructure, and reasoning capabilities to turn AI into real-world intelligence.

Share This Article