Analysis. VAST Data has had a hectic year as the Generative AI training boom has provided an enormous tailwind for its business progress. As a private company, it has no need to publish quarterly trading numbers enabling analysts to track its progress and growth rates. From the hints we are picking up though, it has enjoyed near or actual triple-digit growth rates throughout the year, with a steepening revenue curve as neocloud suppliers like CoreWeave ramp up their storage infrastructure 12 to 24 months behind their GPU compute infrastructure.
CoreWeave, for example, gave it a $1.17 billion order in November, for hundreds of petabytes of VAST Data capacity. This follows several previous orders from CoreWeave. It’s also supplying its storage and software to the Lambda (Lambda Labs as was), Crusoe and Core42 neoclouds as well as xAI’s Colossus systems. Some 98 other customers are listed on its website
It’s hyper-growth and it’s software; the margins are way past 50 percent and we understand VAST is profitable, with a >$1 billion annual revenue run rate, possibly much higher. The company is a phenomenon. The following comments are based on our understanding after talking with VAST execs and trying to understand the nature of the beast.
It seems quite ready for an IPO. A main reason for this would be recognition by large, publicly-owned enterprise customers that it is one of them, and not just a kind of weird, superfast-growing startup with a great reputation. VAST believes that winners buy from winners, and followers buy from the suppliers that sell to winners. In any market sector it enters, it aims to make the leading businesses there its customers, and then the others will surely follow.
It wants its high-level execs to talk to potential customer high-level execs against a background of mutual recognition and awareness. Every such customer exec knows who Dell is, understand’s IBM’s place in the world, and can locate HPE in their IT industry mind-map. VAST wants that status too, that ready access to customer exec mindshare, and believes an IPO would provide it.
VAST has vast confidence in the superiority of its offerings, and believes its filesystem-based competitors are mis-judging the nature of what they are dealing with. They are adding disaggregated controller and storage architectures to their products, along with parallel filesystem access, and stressing performance.
For example, VDURA CEO Ken Claffey says, ”We can beat VAST on performance.” But VAST stresses that, although performance is important, it’s not the vital key. Its disaggregated architecture enables any controller to talk to any drive and that is due to its metadata organisation and infrastructure, provides huge scale, and is likely unique. Its architecture can scale ouit to exabyte levels and do so reliably.
VAST is also focussing more on unstructured object data, and thinks that competitors adding parallelism to filesystems are missing the point. Objects are going to be the main data elements in AI inference and training going forwards, not files.
Its DASE (Disaggregated Shared Everything) architecture underpins its AI-focussed software stack, with DataSpace, DataBase, DataStore, DataEngine, AgentEngine, and InsightEngine layers forming what it calls its AI OS. Has it finished its main development work? No, it says, as interacting AI agents need more infrastructure plumbing, citing the A2A (Agent-to-Agent) protocol and AAIF (Agentic AI Foundation) software stack as areas where it can provide such plumbing.
We’re thinking of agent-level metadata management to help with A2A and AAIF operations; an agent state engine as it were.
VAST has largely grown through organic software development, making the Red Stapler acquisition to get tuck-in technology. This prompted a legal claim from NetApp, where the Red Stapler people previously worked.
NetApp’s lawsuit, case 6:25cv-02130, alleging its now ex-SVP and CTO Jonsi Stefánsson stole NetApp iP and took it to VAST Data when it bought his Red Stapler company, and hired him to be its cloud solutions general manager. It alleged Stefánsson “actively worked against the company that employed him, secretly starting a conflicting business and developing a competing product that he eventually sold to a direct rival company.” Meaning VAST Data.
Our understanding is that this lawsuit is likely to be settled quite soon, and VAST may well make more tuck-in acquisitions.
VAST, to our mind, is like a converged AI software infrastructure company. Whereas it’s disaggregated its hardware components, it has aggregated the AI OS software components, appliance-style. Competitors like Dell and HPE use third-party software elements, such as vector data bases, event brokers and so forth, but VAST has developed its own and is able to say to customers that buying AI stack software from one place, ready-integrated, is simpler.
If and when the long-awaited AI inferencing storage buying increase starts to happen, then VAST can hope to pick up another surge in business. We think that VAST has quite some way to run yet, before its founders run out of innovation steam, and the company throttles back to incremental gains and low double-digit growth rates, or worse.
