The $30B AI infrastructure layer the AI cake is built on
Key points
Key takeaways from a YouTube explainer by technology commentator Natalie Fratto on Vast Data (October 2026):
The $30B blind spot. VAST Data sits in the middle of Jensen Huang's five-layer AI cake (energy, chips, infrastructure, models, applications, last valued at $30 billion — tripling in about 2 years — and profitable, an achievement that's incredibly rare at that growth rate.
Rule of 40. Growth rate plus profit margin: 40 is elite, top public software companies average 61, and VAST posted 228 last year — territory that's quite rare for an infrastructure company.
The infrastructure layer. The company built what it calls an AI operating system serving both massive-scale training and the split-second demands of inference, consolidating data that would otherwise live in disparate systems stored for later query.
Real time, not batch. Vast catches, structures, and fires data out to models for embedding in real time —- a security system catching a threat, a support call that already knows your history, a factory line stopping a bad part before it ships, all near instantly.
Who runs on it. Vast works with Nvidia, CoreWeave, Mistral, and SpaceX AI, as well as data-centric organisations like NASA, Pixar, and the NHL.
The flip for regulated data. Running leading commercial AI models today usually means shipping data to somebody else's cloud; for banks, hospitals, and defense contractors whose data can't leave the building, the next chapter of VAST AIOS flips it — bringing the AI to customers' own hardware, fully encrypted.
Read more: YouTube — Natalie Fratto
