The first obstacle was a file protocol. The National Hockey League wanted a platform for its production operation. VAST Data had storage technology, but it lacked SMB, the protocol the league needed for desktop access. This is an awkward detail for a company selling a new way to handle data: the customer’s computers still have to open the files.
- VAST combines storage, databases and AI processing in one software platform.
- Its architecture lets compute and storage capacity grow independently.
- The NHL, CoreWeave and Hyundai’s 42dot show three different reasons to consolidate data.
The league and VAST worked together. According to the NHL’s customer account, SMB reached production three months later. The archive followed, then a deployment across all 32 arenas. A missing capability became part of the product available to other customers. That sequence is more revealing than an abstract promise about artificial intelligence.
First, let the editor open the file
The NHL’s collection includes more than 550,000 hours of footage and 25 million images. Keeping that material safe is one job. Making it useful to an editor is another. A clip sitting in an archive can preserve history perfectly while being inconvenient to find, retrieve and put into a programme.
By April 2026, VAST described a shared namespace connecting the arenas with the league’s New York office. Editors could access footage without waiting for the old overnight transfers. The distinction matters: the useful outcome is a shorter interval between something happening and somebody being able to work with it.
AI adds another possible use for the archive. Recognising players, tagging scenes and searching footage can turn a collection into something easier to interrogate. Those applications depend on accessible data. The NHL’s experience suggests a sensible order: improve the working infrastructure, then expand what the organisation can do with it.
The expensive computer is waiting
Renen Hallak and his fellow founders started VAST in 2016. Hallak had led research and development at flash-storage company XtremIO. Co-founders Shachar Fienblit, Jeff Denworth and Alon Horev brought experience across storage and distributed systems. VAST emerged publicly in 2019 with Universal Storage and $80 million in disclosed funding.
Its architectural answer is DASE, short for Disaggregated Shared-Everything. In practical terms, the processing resources and storage capacity are separate. Compute nodes access a shared pool of data over a fast network. A customer needing more processing can add compute; one needing more room can add capacity.
Conceptual diagram of VAST’s DASE architecture.
That separation is useful when workloads grow unevenly. More data does not always require proportionally more processors. More demanding analysis does not always require proportionally more storage. VAST’s expertise lies in coordinating those resources while presenting data through interfaces existing applications understand, including NFS, SMB and S3.
Storage was the entrance
In August 2023, VAST introduced a broader Data Platform. DataStore held the data; DataBase added transactional and analytical database capabilities; DataSpace connected locations. DataEngine was announced as the processing layer. The company was moving into the work performed around stored information, extending its relationship with the customer beyond keeping bytes available.
Today, the company calls that foundation the AI Operating System. DataEngine supports event-driven pipelines and Python functions. DataBase supports tables and vector search. DataSpace gives distributed sites a common namespace. Together, these services address the repeated movement and preparation of data between separate tools, a familiar source of delay in AI projects.
Its place in the market depends on what a buyer needs. DDN, WEKA, Pure Storage and IBM compete for demanding storage workloads. Buyers can also assemble cloud storage, databases, search and pipeline services themselves. VAST’s pitch is consolidation. The decision turns on whether sharing one foundation reduces enough operational work to justify changing systems.
The bill needs its own labels
VAST sells enterprise software subscriptions. Its Gemini model, introduced in 2021, separates hardware purchases from software licensing and lets subscriptions move to replacement hardware. The published model licenses usable capacity. The intended benefit is a cluster that can evolve component by component, with fewer forced software repurchases during hardware refreshes.
The public numbers describe different things. CoreWeave’s November 2025 agreement with VAST was valued at $1.17 billion. That is a commercial agreement, not a universal deployment price. In April 2026, VAST announced a $30 billion valuation and approximately $1 billion in combined primary financing and secondary share sales.
VAST also reported more than $4 billion in cumulative bookings and over $500 million in committed annual recurring revenue. Bookings measure commitments; CARR describes contracted recurring business. Neither should be casually relabelled annual recognised revenue. These distinctions become especially useful when large infrastructure contracts make a private company’s growth look deceptively simple.
The customer becomes the investor
On October 13, 2026, Hyundai and Kia announced a strategic investment after production use by 42dot, their autonomous mobility operation. Separate ADAS and language-model platforms had fragmented data and duplicated effort. The company’s account describes consolidation onto VAST and a 2:1 reduction in sensor and video data.
The product is also expanding toward sensitive workloads. September’s DataEnclave announcement described confidential AI execution using NVIDIA technology, intended to protect customer data and model intellectual property during processing. Hallak’s September interview supplied a compact explanation for this direction:
“Models are the new data.”Renen Hallak, September 2026
The transferable lesson is to start with a blockage somebody can name. Count copies, transfers, waiting time and manual handoffs. Test a representative workload. A small application with modest data needs may gain little from this infrastructure; an organisation constrained by incompatible tools or slow data access has a clearer reason to investigate.
The NHL’s story began with opening files. VAST’s ambition now reaches AI agents and model execution. Between those two points sits the discipline buyers should copy: make the existing job work, prove the improvement, and earn the next job.