Technology

VAST Data

A disaggregated all-flash data platform providing unified access via S3, NFS, and SMB protocols, optimized for AI and deep learning workloads with consistent low latency.

5 connections 2 resources 2 posts

Summary

What it is

A disaggregated all-flash data platform providing unified access via S3, NFS, and SMB protocols, optimized for AI and deep learning workloads with consistent low latency.

Where it fits

VAST Data targets the convergence of AI/ML workloads and object storage. Its all-flash architecture eliminates the cold scan latency that plagues spinning-disk object stores, while the S3 interface maintains ecosystem compatibility.

Misconceptions / Traps
  • Not just object storage. VAST is a unified data platform where S3 is one of multiple access protocols. Evaluating it solely as an S3 alternative misses its multi-protocol value.
  • All-flash means higher per-GB cost than HDD-based object stores. The value proposition is performance per dollar, not lowest cost per GB.
Key Connections
  • implements S3 API — S3-compatible interface
  • solves Cold Scan Latency — all-flash eliminates seek latency

Definition

What it is

A disaggregated, all-flash data platform providing S3-compatible object storage alongside file (NFS/SMB) and database access on a single unified architecture optimized for AI workloads.

Why it exists

AI/ML workloads need both high-throughput object storage and low-latency file access. VAST eliminates the need for separate storage systems by unifying all protocols on an all-flash platform with S3 compatibility.

Primary use cases

AI/ML training data platform, unified NFS+S3 storage, high-performance analytics, GPU-direct data access.

Recent developments

Latest signals
  • Series F at $30B valuation — $1B round led by NVIDIA's AI-storage bet. Per VAST Data's press release on the Series F financing (April 22, 2026), VAST closed a $1B Series F at a $30B valuation. NVIDIA participation flagged across coverage as the marquee AI-infrastructure signal — VAST's positioning as the unified storage substrate for AI-training data loops is now backed by the GPU vendor with the most direct interest in storage stops being the bottleneck.
  • "Collapsing the Stack" strategy — own the AI data loop end-to-end. VAST's strategic frame for 2026 is consolidating the AI-data-pipeline tiers (object + file + database + lineage) onto a single all-flash platform — competing not just with object stores like FlashBlade and StorageGRID but with the broader "AI data lake + feature store + RAG infrastructure" surface that has historically been a multi-vendor patchwork. Per VAST — N=1: How VAST Built the De Facto Data Layer for AI.
  • The financials behind the $30B (June 2026 detail) + the CoreWeave anchor. Coverage of the Series F put hard numbers on the valuation: $4B+ cumulative software bookings, $500M+ committed ARR, ~90% gross margins, and 300%+ net revenue retention — plus a landmark $1.17B commercial agreement with CoreWeave that makes VAST the default storage supplier to GPU-heavy neoclouds. Per Calcalist — VAST valuation soars to $30B and Futuriom — VAST hits $30B.
  • From storage vendor to "AI Operating System" — native on BlueField-4. VAST's DASE (Disaggregated Shared Everything) architecture — stateless compute over a shared NVMe-oF flash pool — now bundles storage + a transactional database + a compute engine as one platform. Running a VAST CNode natively on NVIDIA BlueField-4 (STX reference architecture) enables zero-copy NVMe→GPU-VRAM paths via GPUDirect Storage + RDMA; VAST-reported benchmarks show a 90% inference-efficiency gain and a 20× Time-to-First-Token improvement when offloading KV caches to the tier. This is the same DPU-offload pattern the ROS2 ACM paper validates academically. Per VAST — N=1 data layer for AI.

Connections 5

Outbound 4
implements1
Inbound 1
competes_with1

Resources 2

Featured in