Nashawaty: 86% of enterprises pick data over compute for AI

The event's framing starts from a shift that Dell and theCUBE Research describe as already underway: enterprises have spent heavily on models and accelerated computing, but the harder problem is getting distributed, messy and sensitive data ready for systems that need reliable context at speed. Paul Nashawaty's numbers — 86% prioritizing data unification over compute, 64% of AI teams citing insufficient storage throughput — are presented as evidence that this is where the bottleneck now sits, not in model selection.

A second thread runs through the practitioner interviews. Dell's Vrashank Jain describes agents as unpredictable in the data they may need, reaching across SharePoint, OneDrive, SaaS applications and legacy systems, and firing roughly 10 times more queries than before. Elastic's Sri Desikan points to a related architectural move: pre-building context so models don't repeatedly search back-end systems, using vector search, hybrid search and re-ranking, with the aim of minimizing token costs and maximizing accuracy. John Furrier frames this emerging context layer as connective tissue between models and enterprise data, and draws a line between a good demo and production-grade discovery and low latency.

Security and governance sit alongside that. Krista Case of theCUBE Research says the challenge expands beyond protecting models to protecting the distributed data that gives AI context and value, and that cyber resilience will depend on knowing where critical data lives and controlling how it is accessed. Whether the Dell AI Data Platform's architecture — bringing AI closer to enterprise data while keeping controls around security, governance and sovereignty — becomes the pattern enterprises adopt is likely to hinge on how these production requirements are met as deployments scale.