VAST Data valuation hits $30 billion as AI inference demand scales

The massive valuation confirms that infrastructure constraints have shifted from raw compute availability to the data movement layer. As inference workloads multiply, the KV cache—a key-value store for intermediate model states—becomes a critical bottleneck for low-latency performance in disaggregated architectures. VAST's entry into the neocloud market, offering a full catalog of data services on top of GPU infrastructure, poses a direct threat to hyperscalers by promising 98% gross margins for its partners. Watch for whether VAST targets an IPO in late 2026, as secondary market activity suggests a liquidity event may be nearing.

The rise of KV cache storage as a standalone infrastructure priority was underscored by Nvidia’s January 2026 launch of the CMX Context Memory Storage Platform. Per TrendForce (June 2026), this platform uses BlueField-4 DPUs to manage context tiers between local SSDs and shared storage, addressing an explosion in token output that now exceeds 30,000 tokens per question. This shift aligns with the emergence of agentic AI, where models operate in 'thinking' mode over extended periods, requiring persistent memory systems that do not flush context during inference decode cycles.

Technically, the AI stack is undergoing a structural collapse as traditionally separate layers like storage, databases, and event brokers converge. At the VAST Forward 2026 conference in February, the company debuted its fully CUDA-accelerated AI operating system, which runs directly on Nvidia-powered CNode-X servers. According to Futuriom (April 2026), this integration allows VAST to maintain 90% gross margins while surpassing $4 billion in cumulative software bookings. The company’s revenue roughly tripled annually between 2023 and 2025, reaching an estimated $2 billion annual recurring revenue (ARR) threshold by the start of 2026.

VAST’s market position is cemented by its role as Nvidia's only SuperPOD-certified storage partner. Per Calcalist (April 2026), the recent Series F round was uniquely heavy on secondary capital, with over $500 million going to early investors and employees. This structured liquidity provides a bridge for the company as it scales its 'AI factory' model with partners like Cisco and Supermicro. Competitors such as DDN have also signaled IPO ambitions in early 2026, indicating a broader market rush to capture the spend moving from training clusters to production-grade inference stacks.