Oracle’s Larry Ellison Sounds Alarm on AI Commoditisation, Urges Private Data Focus

Oracle co-founder and CTO Larry Ellison has warned that today’s leading AI models are rapidly becoming commoditised because they are trained on the same public internet data.

He argues that the next major breakthrough in AI will come from systems that securely leverage private enterprise data, not generic public sources. 

Larry Ellison, the co-founder and CTO of Oracle, has directly challenged the current trajectory of artificial intelligence development, saying that major AI models, including ChatGPT, Google’s Gemini, xAI’s Grok and Meta’s Llama, share a fundamental weakness: they all train on the same publicly available internet data. Ellison told investors that this shared foundation makes cutting-edge AI systems increasingly similar, turning them into commodity products with limited real differentiation.

Ellison emphasised that the “real value” for AI’s future lies in giving models secure access to private, proprietary data held by enterprises, which he believes will unlock deeper insights and far greater economic potential than public-data-only systems. 

To support this shift, Oracle has sharply increased its commitment to AI infrastructure, with Ellison revealing plans to invest nearly $50 billion in capital expenditure to build large-scale AI computing capacity and enterprise-focused platforms. He said these investments aim to enable AI systems to reason over private data securely, rather than rely solely on generic public information.

Ellison’s message highlights why expertise in data science, artificial intelligence and agentic AI is becoming vital. Professionals with these skills can design AI systems that integrate diverse data sources, build advanced reasoning capabilities and create custom intelligent agents that go beyond generic outputs and drive real-world business value.

What Ellison Sees as AI’s Fundamental Flaw

Ellison argues that most of today’s popular AI models are converging into the same outcome because developers train them on the same pool of publicly available internet data, websites, articles, forums and other open-source content. This shared data foundation means that models tend to produce very similar answers, regardless of their developer or claimed innovation.

“In essence, all these models are studying from the same book,” Ellison said, adding that this trend risks making advanced AI indistinguishable and vulnerable to a race to the bottom on price and features, rather than quality and differentiation.

Why Public Data Alone Isn’t Enough

Ellison said that relying solely on public internet data limits AI’s usefulness because such data often lacks depth, context and exclusivity. While training on public sources can build general-purpose knowledge, it doesn’t help companies solve highly specific business problems that depend on confidential or domain-specific information.

He pointed out that many AI systems today can sound intelligent but still fail to deliver actionable insights because they cannot reason with sensitive datasets that businesses hold such as financial records, operational workflows or customer behaviour.

The Case for Private, Proprietary Data

Ellison sees the next phase of AI innovation coming from models that reason over private enterprise data in secure environments. He believes this shift could surpass the economic value generated by the current focus on GPUs, data centres and public-data-trained models.

Oracle has argued that because much of the world’s most valuable data already resides in its enterprise databases, the company is uniquely positioned to drive this transition. Its AI Data Platform aims to let AI models query private data in real time using techniques like Retrieval-Augmented Generation (RAG), without compromising security or privacy.

Oracle’s Strategic Investment in Enterprise AI

To support its vision, Oracle has dramatically increased its investment in AI infrastructure. The company now projects approximately $50 billion in capital expenditures for the fiscal year  up from earlier estimates to build high-performance AI systems, including a 50,000-GPU supercluster powered by AMD chips and the OCI Zettascale10 supercomputer linking hundreds of thousands of Nvidia GPUs.

Oracle is positioning these systems to power enterprise AI applications that operate on private data, underpinning workflows in finance, supply chain, healthcare and other data-intensive industries.

Conclusion

Larry Ellison’s warning is a reality check: size doesn't matter if everyone is building the same thing. The "commodity" era of AI is forcing a pivot toward specialized, private intelligence. As Big Tech models become interchangeable, the real power shifts to those who control the private data pipelines. The future belongs to the "electronic brains" that know your business secrets, not just what's on Wikipedia.