https://youtu.be/SfN5LmWWgss?si=thKy5vs40kJPM3Vc
Larry Ellison’s idea is actually quite profound. He believes that the future of AI is not determined primarily by who has the largest language model, but by who has the richest private data.
He often says that today’s frontier models—whether from OpenAI, Google, Anthropic, xAI, or Meta—are all trained largely on the same public Internet. As a result, they risk becoming commodities. The real competitive advantage comes from connecting those models to proprietary enterprise data.
His vision can be summarized as follows.
The Internet is AI’s kindergarten
Current LLMs learn from
- books
- Wikipedia
- websites
- GitHub
- public papers
That produces a model with broad general knowledge.
But it knows almost nothing about
- Toyota’s manufacturing process
- Pfizer’s research data
- JPMorgan’s transactions
- your hospital’s patient records
- your company’s contracts
Those are the data that actually run businesses.
Ellison argues that this private data—not public web text—is where the greatest economic value lies.
Oracle’s database becomes AI’s memory
Traditionally,
Database
↓
Application
↓
Human
The database simply stored rows and columns.
Ellison wants to change that into
Enterprise Database
↓
Vector Database
↓
Foundation Model
↓
AI Agent
↓
Human
In this architecture, the database is no longer passive storage.
It becomes the AI’s long-term memory.
Why “vectorize everything”
This is the key idea.
Traditional databases store facts like:
AI doesn’t naturally understand tables like these.
Instead, Oracle converts records, documents, emails, images, manuals, and contracts into vector embeddings—numerical representations of meaning—so the AI can retrieve semantically related information rather than relying on exact keywords. Oracle has built vector data types and vector search directly into its database engine for this purpose.
Ellison’s phrase is essentially:
“Vectorize all your data.”
Then AI can understand
- meaning
- similarity
- context
instead of only exact matches.
Database becomes a “brain”
Ellison is trying to transform the database from
Storage
into
Semantic Memory
That resembles long-term memory in cognitive science.
An LLM supplies
- language
- reasoning
- planning
The Oracle database supplies
- memory
- facts
- organizational knowledge
- history
Together, they create an enterprise AI system.
AI agents instead of chatbots
Ellison has repeatedly emphasized that the future is not chatbots but AI agents.
Imagine asking:
“Why were sales down in Taiwan last quarter?”
Instead of generating a plausible answer,
the agent would
- search the ERP
- examine invoices
- compare warehouse inventory
- read emails
- analyze customer complaints
- generate charts
- recommend actions
using the company’s own data.
Oracle is designing its database platform to make this kind of retrieval and reasoning straightforward.
Why Oracle thinks it has an advantage
This is classic Ellison thinking.
OpenAI owns
models.
Oracle owns
data.
More precisely,
Oracle is already the custodian of enormous amounts of mission-critical enterprise information across governments, banks, hospitals, manufacturers, and retailers. Ellison argues that this makes Oracle especially well positioned for AI inference built on private data.
My comment
I think Ellison is identifying one of the most important shifts in AI.
The first phase of AI was:
Train on the world’s public knowledge.
The next phase is:
Reason over the world’s private knowledge.
That is a much larger market.
Interestingly, this also connects to one of the themes we’ve been discussing over the past year.
You have often said that AI is learning to think through language.
Ellison is adding another ingredient:
Language alone is insufficient.
An intelligent agent also needs memory.
In neuroscience, we might say:
- the LLM resembles a highly capable cortex for language and reasoning;
- the enterprise database resembles a semantic and episodic memory store.
Neither is enough by itself. Intelligence emerges from the interaction between reasoning and memory.
I suspect that, over the next decade, the most valuable AI systems will not be the ones with the very largest models, but those that can securely and accurately combine powerful reasoning with deep, organization-specific knowledge. In that sense, Ellison’s strategy is less about building a smarter chatbot and more about building AI that can function as a knowledgeable colleague within a business.
