In a bold statement that challenges the current trajectory of artificial intelligence, Oracle's founder and Chief Technology Officer, Larry Ellison, has pinpointed a major flaw holding back leading AI models from Google's Gemini to OpenAI's ChatGPT. According to Ellison, these models have become commoditized because they are all trained on the same publicly available internet data.
The Public Data Trap: Why AI Models Lack Differentiation
Speaking during Oracle's fiscal second-quarter earnings call for 2026 in December, Ellison acknowledged that training foundational models on public data has created what he termed "the largest and fastest-growing business in human history." However, he argued that this approach has led to a critical limitation: a lack of real differentiation between competing AI services. The future of AI's value, he contends, lies in a crucial second phase.
"For these models to reach their peak value, you need to train them not just on publicly available data, but make privately owned data available for those models as well," Ellison stated. He estimates this next evolution, focused on secure, private data, will prove "even larger and more valuable" than the current boom in GPU and data center investments.
Oracle's $50 Billion Bet to Become the AI Backbone
Oracle is aggressively positioning itself to capitalize on this predicted shift. The company's strategy hinges on a key advantage: most of the world's high-value private corporate data already resides in Oracle databases. To leverage this, Oracle's AI Data Platform employs techniques like Retrieval-Augmented Generation (RAG). This allows major AI models to query and reason over private enterprise data in real-time without compromising security protocols.
Backing this vision with substantial capital, Oracle has projected capital expenditures of roughly $50 billion for the full year, a significant increase from the $35 billion estimated in September 2025. The company is also building formidable infrastructure, including a 50,000-GPU AI supercluster with AMD MI450 chips set to launch in Q3 2026, and the OCI Zettascale10 supercomputer connecting hundreds of thousands of NVIDIA GPUs.
The Intensifying Race for Enterprise AI Dominance
Ellison's thesis, however, faces significant challenges and competition. Industry observers note that synthetic data generation could reduce the need for exclusive proprietary datasets. Furthermore, real-time user interaction data from consumer applications might ultimately prove more valuable than static enterprise records.
Cloud rivals like Amazon Web Services (AWS), Microsoft Azure, and Google Cloud are also racing to develop comparable enterprise AI capabilities. Oracle counters that its entrenched position within enterprise databases provides a unique strategic edge. The demand is evident: by late 2025, Oracle's cloud backlog had surpassed $500 billion, driven largely by AI-related demand.
The battle for the future of enterprise AI is clearly heating up, with Oracle making a monumental financial and technological wager that the key to unlocking AI's true potential lies not on the open web, but within the secure vaults of private corporate data.
