Oracle cofounder and CTO Larry Ellison has offered a blunt assessment of the modern AI race, arguing that nearly every major artificial intelligence model suffers from the same underlying problem. Speaking during Oracle’s fiscal Q2 2026 earnings call in December, Ellison said that today’s most prominent AI systems are trained on essentially identical datasets, which is rapidly stripping them of meaningful differentiation.
According to Ellison, models from companies such as OpenAI, Google, Meta, Anthropic, and xAI are all built on the same foundation of publicly available internet data. That shared training source, he says, is turning what once felt like cutting-edge innovation into a commodity. As Ellison put it, when everyone is using the same data, the resulting models inevitably start to look and behave the same.
This observation cuts to the heart of a growing concern in the AI industry. While products like ChatGPT, Gemini, Grok, and Meta’s Llama differ in branding and fine-tuning, their core knowledge largely overlaps. Ellison argues that this makes it increasingly difficult for any single provider to maintain a lasting competitive edge purely through model improvements.
In Ellison’s view, the next major phase of AI growth will not come from building ever-larger foundation models. Instead, the real value will be unlocked by enabling AI systems to work securely with private enterprise data. He believes this second wave of AI adoption could ultimately be larger and more profitable than the current boom driven by GPUs, massive data centres, and consumer-facing chatbots.
Oracle is positioning itself squarely around this thesis. The company argues that most high-value data in the world already sits inside enterprise databases, many of which are managed by Oracle. By allowing AI models to interact with that proprietary data while keeping it secure, Oracle believes it can offer something far more defensible than generic public models. Its approach relies heavily on techniques such as retrieval-augmented generation, which allows AI systems to query private datasets in real time without exposing sensitive information or retraining the underlying model.
This strategy is also driving Oracle’s aggressive investment in AI infrastructure. The company now expects to spend around $50 billion in capital expenditures for the full year, up sharply from an estimate of $35 billion just months earlier. Much of that spending is focused on building large-scale AI compute capacity to serve enterprise customers.
At Oracle AI World in October, the company announced several high-profile infrastructure projects, including a 50,000-GPU supercluster based on AMD MI450 chips planned for launch in the third quarter of 2026. It also revealed details of the OCI Zettascale10 supercomputer, designed to connect hundreds of thousands of NVIDIA GPUs. By late 2025, Oracle’s cloud backlog had reportedly crossed $500 billion, with AI-related demand accounting for a significant portion of that growth.
Still, Ellison’s vision is not without challenges. Advances in synthetic data generation could reduce the importance of exclusive proprietary datasets over time. At the same time, rivals such as Amazon Web Services, Microsoft Azure, and Google Cloud are racing to offer their own enterprise-focused AI platforms. Whether Oracle’s long-standing dominance in enterprise databases will translate into a lasting advantage in AI remains an open question.
For now, Ellison’s comments highlight a critical tension in the AI industry. As models trained on public data converge, differentiation is becoming harder to sustain. If he is right, the future of AI will be shaped less by who builds the smartest chatbot and more by who can safely and effectively connect AI 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!
