Oracle Vectorizes Its Customers Data (www.theregister.com)

At Oracle AI World, CTO Larry Ellison announced that Oracle has “vectorized” its customers’ data — converting records across Oracle and third‑party databases and clouds into embeddings so they’re directly usable by its AI “reasoning” models. Ellison said this pipeline powers predictive agents that, for example, forecast which products a customer is likely to buy in the next six months and automatically generate targeted outreach (including suggested customer references). The pitch is operational: make private enterprise data searchable and actionable for retrieval‑augmented reasoning models and agentic workflows. The move matters because it accelerates adoption of embedding + LLM stacks inside core enterprise systems while raising technical, commercial and ethical flags. Technically, it centralizes high‑dimensional vector stores tied to reasoning models and agent generators, increasing reliance on massive retrieval and GPU inference infrastructure. Ellison also outlined an audacious datacenter buildout (an Abilene, TX site he said could host ~500k Nvidia GPUs and consume ~1.2 GW, across 1,000 acres), and referenced major financial bets and projects (AI backlog, debt and partnerships). Implications include new monetization and auditing vectors for vendors, consent and privacy concerns as customer records are repurposed, and significant environmental and regulatory scrutiny given the scale of compute and claimed synthetic‑biology/CO2 projects.