Vast Data

VAST Data was founded in 2016 by Renen Hallak and Jeff Denworth, along with several other co-founders from the enterprise storage industry. Renen Hallak previously held senior engineering roles at XtremIO, the all-flash storage startup that EMC acquired in 2012 for approximately $430 million. The experience at XtremIO — building high-performance flash arrays inside a legacy storage giant — gave Hallak a precise understanding of where existing architectures hit their limits, and VAST Data was built explicitly to circumvent those limits using the economics of QLC NAND flash and NVMe interconnects that had become cost-viable by 2016.

VAST Data is headquartered in New York City (Manhattan), with its primary R&D center in Tel Aviv. Additional offices exist in San Jose, California, serving North American sales and business development, and the company has sales presence in London, Munich, Singapore, and Tokyo as it has expanded to serve enterprise and research customers globally.

VAST Data remains a private company as of mid-2025. In February 2024 the company closed a funding round of $118 million at a post-money valuation of approximately $9.1 billion, making it one of the most highly valued private infrastructure software companies in the world. Total capital raised since founding exceeds $400 million across multiple rounds, including an $83 million Series D in 2021 when the valuation was approximately $3.7 billion, effectively more than doubling the valuation in the three years between those two raises.

VAST Data employs approximately 800 to 900 people globally as of 2024, with roughly 300 to 400 of those based in Israel. The Israel headcount is concentrated almost entirely in R&D functions — the storage engine team, the metadata systems group, distributed systems engineering, and QA. The remainder of the global workforce is split between North America (largely sales, marketing, and customer success) and international offices in Europe and Asia-Pacific.

The core product is a unified AI storage platform — a hardware-software appliance and software-defined stack that delivers high-throughput, low-latency storage for GPU-heavy AI training and HPC workloads, combining NVMe flash hardware with a novel disaggregated architecture.

The single most consequential product event in the last 12 months was the launch of VAST DataEngine in 2024, which added a native vector database and semantic search capability to the platform, directly positioning VAST as a storage-layer competitor to standalone vector databases like Pinecone and Weaviate in Retrieval-Augmented Generation (RAG) pipelines. This move transformed VAST from a pure storage play into a broader data platform company.

VAST Data is not a subsidiary of any larger entity. It is independently operated and has not been acquired.

The primary product line is the VAST Data Platform, comprising four distinct named components: VAST DataStore (the storage layer), VAST DataBase (a SQL query engine), VAST DataEngine (a vector database and semantic search layer), and VAST DataSpace (the orchestration and lifecycle management control plane). These four components are sold together as an integrated platform, though DataStore remains the foundational product that generates the majority of revenue.

The domain-specific problem VAST solves is storage I/O bottleneck in large-scale GPU training clusters. When organizations assemble GPU clusters consisting of hundreds or thousands of NVIDIA H100 or A100 cards — whether on-premises or in a colocation facility — the aggregate I/O bandwidth those GPUs demand during training runs frequently saturates or overwhelms conventional storage systems. Traditional NAS (like Dell EMC Isilon), object storage (like Ceph or S3), and even older all-flash arrays (like those based on scale-up NVMe) cannot sustain the sustained read throughput of 400–800 GB/s or higher that large DGX SuperPOD configurations require. VAST's disaggregated shared-everything architecture addresses this specific bottleneck.

VAST's buyers are large enterprises and research institutions: financial services firms running quantitative models and risk analytics (hedge funds, tier-one investment banks), academic and government research laboratories (national labs in the United States and Europe), life sciences organizations doing genomics and drug discovery compute, media and entertainment companies doing AI-assisted rendering, and large technology companies building their own AI infrastructure. The buying role is typically the VP of Infrastructure, the Head of AI/ML Engineering, or the CTO. VAST does not sell to small or mid-market companies — minimum deployment costs begin in the range of several hundred thousand dollars.

The sales model is entirely sales-led, with enterprise deal cycles that range from 6 to 18 months. VAST works with channel and OEM partners including Dell Technologies and HPE, which resell VAST-based configurations to their own enterprise customers. NVIDIA recommends VAST as a validated storage partner for DGX BasePOD and DGX SuperPOD reference architectures, which gives VAST significant pipeline influence. There is no self-serve tier, no free community edition, and no public pricing list.

Pricing is structured as enterprise contracts, negotiated on the basis of raw capacity (petabytes), number of storage nodes, performance tier, and support level. The company has not disclosed precise ARR figures, but commentary from investors in 2024 implies ARR in the range of several hundred million dollars.

VAST's technical moat rests on three concrete pillars. First, the Disaggregated Shared Everything (DASE) architecture, which is the subject of multiple patents filed since 2017, separates compute and storage while maintaining shared-everything semantics — avoiding the hot-spot and rebalancing problems that plague scale-out shared-nothing architectures. Second, the Global Namespace capability allows a single unified namespace to span multiple petabytes without directory sharding, solving a problem that has plagued Lustre and GPFS deployments for decades. Third, the element-level metadata encoding scheme provides RAID-like protection at the metadata block level, a technical detail VAST has published in papers presented at the USENIX FAST conference.

The engineering organization builds primarily in C++ and Rust for the core storage engine and kernel-level components, Python for metadata services and orchestration tooling, and integrates with NVIDIA's GPUDirect Storage API to minimize CPU overhead in GPU-to-flash data paths. The company participates in the MLCommons MLPerf Storage benchmark consortium and has submitted benchmark results showing top-tier throughput for AI training workloads.

VAST DataStore is the foundational product and the one against which all competitors are compared. It presents a unified file, object, and block interface — simultaneously accessible over NFS v3 and v4.1, SMB 3.0, S3-compatible API, and NVMe-over-Fabrics (NVMe-oF) — over an all-QLC NVMe flash backend. The compression and inline deduplication algorithms VAST uses are specifically tuned for AI training data (model checkpoints, activation tensors, datasets of images and text), achieving reduction ratios that the company claims make all-flash total cost of ownership competitive with or lower than hybrid spinning-disk systems for the specific use case of model training.

VAST DataBase, launched in 2023, is a SQL query engine that runs directly against data stored in DataStore without an ETL step. It exposes a PostgreSQL-compatible wire protocol and supports Apache Arrow Flight for high-bandwidth data transfer to analytical frameworks. DataBase means that Spark, DuckDB, Tableau, and other analytics tools can query petabyte-scale datasets that physically live on the storage tier — effectively turning VAST into a competitor to on-premises data lakehouse platforms.

VAST DataEngine, launched in 2024, embeds a vector database directly into the storage platform. It supports approximate nearest-neighbor (ANN) vector search against embedding vectors stored natively, meaning organizations building RAG pipelines for large language model applications do not need to run a separate vector database product like Pinecone, Weaviate, or Milvus. This is the most recently launched product and the one that signals VAST's strategic ambition beyond pure storage.

VAST DataSpace, also launched in 2024, is the unified control plane that manages data lifecycle policies, automated tiering to S3-compatible external object stores, namespace management, and access control across the entire platform. It functions as the management layer that ties DataStore, DataBase, and DataEngine into a single operational experience.

All four products were developed internally — VAST has not acquired a company to date. The flagship is DataStore by revenue. DataBase, DataEngine, and DataSpace are positioned as platform add-ons that increase land-and-expand revenue from existing accounts. No VAST products have been sunset to date. VAST received FedRAMP Authorization in 2022, which opened U.S. federal government and Department of Defense procurement channels. The company has also completed SOC 2 Type II audits, which are required for regulated financial services customers.

The most direct competitor to VAST Data for AI training storage workloads is IBM Storage Scale (formerly Spectrum Scale, formerly GPFS), which has been the de facto standard in national laboratory and government research HPC environments since the early 2000s. IBM Storage Scale offers decades of production hardening and deep integration with IBM's HPC middleware stack (Spectrum LSF, Spectrum MPI), but is built on an architecture that predates all-flash NVMe and shows scalability limits when serving multi-thousand-GPU clusters at sustained high throughput. VAST has won displacement deals against IBM Storage Scale at several national research labs, and published MLPerf Storage benchmark results in 2023 and 2024 that show VAST outperforming legacy parallel file systems at GPU cluster scale.

NetApp, listed on NASDAQ under the ticker NTAP with annual revenues of approximately $6 billion, is a second significant competitor, particularly through its AFF (All-Flash FAS) A-series and the EF-series NVMe arrays. NetApp has substantial enterprise relationships and a more comprehensive support and services organization than VAST, but its ONTAP operating system was architected for scale-up HA pairs rather than disaggregated scale-out, which limits aggregate throughput for the largest GPU cluster configurations.

WekaIO (operating as Weka) is a third named direct competitor, and a particularly interesting one because it is also an Israeli-American company — headquartered in San Jose with a large R&D center in Tel Aviv. Weka offers a high-performance parallel file system that competes directly with VAST for AI and HPC storage, and both companies frequently appear on the same customer shortlists. The primary distinction is that Weka has focused on the parallel file system layer and does not offer an integrated SQL query engine or vector database, whereas VAST has moved toward a full data platform positioning. Weka raised $135 million in a Series D round in 2023 at a valuation of approximately $1.6 billion.

Pure Storage, listed on NYSE under the ticker PSTG, is a fourth competitor through its FlashBlade//S product line, which offers scale-out all-flash object and file storage. Pure Storage has a significantly larger installed base and revenue (approximately $2.8 billion in fiscal year 2024) than VAST, but FlashBlade//S is positioned as a general-purpose enterprise storage platform rather than a purpose-built AI training appliance.

VAST does not appear in the Gartner Magic Quadrant for Primary Storage, which has historically focused on vendors with broader enterprise market coverage. The company has received favorable coverage in Gartner Peer Insights in the Distributed File Systems and Object Storage category. The company's pricing is positioned as premium — typical enterprise contracts run into the millions of dollars — but it competes on total cost of ownership by demonstrating that compression and deduplication efficiency can make all-flash more economical than spinning-disk hybrid systems over a 5-year lifecycle.

Reported customer wins include Fermilab (the U.S. Department of Energy's particle physics laboratory in Illinois), Nasdaq (the exchange operator, for its data analytics infrastructure), and an undisclosed number of financial services firms that have not been named publicly. The company is gaining share in the AI infrastructure buildout, which represents one of the strongest demand tailwinds in enterprise infrastructure spending in 2023–2025. The principal headwind is the rise of hyperscaler-managed storage services (AWS FSx, Google Filestore, Azure HPC Cache), which threaten on-premises HPC storage as some AI training workloads migrate to cloud. VAST has not made any acquisitions.

VAST Data's Israel operations are based in Tel Aviv. The company has occupied office space in central Tel Aviv since its founding in 2016, and as of 2023 expanded its Tel Aviv footprint as part of a broader headcount growth initiative funded by the Series D and E capital raises. No relocations to other Israeli cities have been announced, and the Tel Aviv office remains the company's single largest global engineering concentration.

Israel is home to approximately 300 to 400 of VAST's roughly 800 to 900 total employees — making Israel the largest single-country engineering hub in the company. The functions based in Israel are dominated by R&D: the storage kernel engineering team (C++ and Rust), the distributed metadata systems group, the networking and protocol engineering team (NFS, SMB, NVMe-oF, S3), the QA and test automation group, and elements of the DevOps and release engineering function. Customer success and sales for the Israel and EMEA regions are also partly based in Tel Aviv.

VAST Data has not announced any workforce reductions in Israel. In contrast to the broad wave of Israeli tech layoffs seen across the sector in 2022 and 2023, VAST continued to hire through those years, supported by the capital it raised in 2021 and 2024. The company announced in early 2024, alongside the $118 million raise, its intention to continue expanding R&D headcount, with Israel specifically named as a growth location.

Renen Hallak, the CEO and primary founder, is Israeli and previously built his career in Israeli enterprise tech at XtremIO in Tel Aviv before that company's acquisition by EMC. Multiple members of the senior engineering leadership team are Israeli, and there is a well-documented pipeline from IDF technology units — including Unit 8200 and Unit 81 — into VAST's engineering organization, consistent with the broader pattern across Israeli enterprise infrastructure companies. Jeff Denworth, the American co-founder who served as CMO until departing in 2023, is not Israeli.

The roles VAST Data most consistently recruits for in Israel include senior C++ and Rust engineers for storage kernel and systems programming, Linux kernel engineers, distributed systems engineers with experience in consensus protocols and fault-tolerant system design, storage protocol engineers with deep knowledge of NFS v4.1 and NVMe-oF, QA automation engineers with systems-level testing experience, site reliability and platform engineers, and technical product managers with HPC or AI infrastructure backgrounds. The company also recruits technical writers in Israel for its engineering documentation.

VAST Data's investor base includes 83North, the Israel-headquartered venture capital firm that also backs other Israeli B2B technology companies, giving the company a notable Israeli institutional investor anchor. Other investors include Fidelity Management & Research, Tiger Global Management, Greenfield Partners, and Next47 (the venture arm of Siemens AG). The company's culture in Israel is reported in hiring circles to offer compensation packages — including equity grants — that are competitive with U.S. hyperscaler offers, a deliberate strategy to compete for senior systems engineers who receive competing offers from Google, Meta, and Apple's Israeli R&D centers.