AI storage: which startup is ahead?

Why is AI storage suddenly becoming such a big market?

AI storage is getting a lot more attention because companies are spending billions on GPU clusters, and those expensive chips are only useful when they can access data quickly enough.

The problem goes beyond storing training datasets. AI systems repeatedly read large files, save model checkpoints, retrieve information, and move data between computing environments. When storage cannot keep up, GPUs spend time waiting instead of processing information.

The amount of money flowing into specialized storage companies shows how important this has become. VAST Data's valuation climbed from $9.1 billion in late 2023 to $30 billion in 2026. WEKA crossed $100 million in annual recurring revenue during 2024. Hammerspace reported that its bookings in the first part of 2026 were already nearly fourteen times higher than its entire previous year. A jump from $9.1 billion to $30 billion is why we lay out what these companies are worth in our AI infrastructure report.

There's also a change in what customers want. Training large language models originally drove much of the demand for specialized AI storage. Now inference, AI agents, autonomous vehicles, and enterprise AI applications are creating different requirements.

A training cluster might need to read enormous datasets at maximum speed. An enterprise running thousands of AI agents needs fast access to frequently changing information spread across multiple systems.

Storage companies are competing for a much bigger role in AI infrastructure, with some trying to become the main data platform behind entire AI operations.

Who is actually leading the AI storage startup market right now?

VAST Data is currently the strongest AI storage startup overall, although WEKA remains a serious competitor in high-performance storage and Hammerspace has an advantage in connecting data across existing systems.

VAST has the most convincing combination of financial scale, major customer contracts, and profitability. Its latest financing valued the company at $30 billion, far above WEKA's last publicly disclosed $1.6 billion valuation.

But the companies are solving somewhat different problems.

VAST sells a unified data platform designed to support everything from large AI training clusters to databases and inference applications. WEKA specializes in extracting maximum performance from demanding computing environments, particularly GPU clusters. Hammerspace helps organizations access information spread across different storage systems without having to move everything into one place.

DDN also belongs in the discussion. Founded in 1998, the company built its reputation in scientific computing and supercomputing long before the generative AI boom. Its reported annual revenue is substantial enough to challenge the newer companies, although calling it an AI startup would be misleading.

For our ranking, commercial adoption carries more weight than product announcements or isolated performance benchmarks. A storage company can have excellent technology without winning the largest contracts.

That's where VAST currently stands apart.

How much bigger is VAST Data than WEKA, Hammerspace and DDN?

VAST Data has the strongest disclosed recurring business among newer AI storage startups, while DDN may already generate more annual revenue than any of them.

In its April 2026 financing announcement, VAST reported more than $500 million in committed annual recurring revenue and over $4 billion in cumulative bookings. It also disclosed positive operating margins and free cash flow.

WEKA's last major published financial milestone came in 2024, when it passed $100 million in ARR after doubling revenue for three consecutive years. That establishes a meaningful business, although the figure is now too old to represent its current size.

DDN provides a useful comparison. According to Forbes reporting in August 2026, its revenue reached approximately $500 million in 2025, up from around $400 million in 2024. Management projected roughly $1 billion for 2026.

Hammerspace is harder to measure. It has reported extraordinary bookings growth but hasn't disclosed enough absolute revenue information to compare its current size reliably with the others.

There is an important accounting distinction. VAST's committed ARR includes contracted recurring business, whereas DDN's annual revenue reflects sales recognized over a completed financial period. The figures shouldn't be treated as interchangeable.

Company	Latest public financial evidence	Valuation reference	What we can conclude
VAST Data	Over $500M committed ARR	$30B, 2026	Largest disclosed recurring commitments among newer challengers
DDN	Approx. $500M revenue in 2025	$5B, 2025	Major established private competitor
WEKA	Over $100M ARR in 2024	$1.6B, 2024	Proven business; newer ARR unavailable
Hammerspace	Nearly 14× prior full-year bookings in early 2026	Above $500M, reported in 2025	Rapid expansion; absolute revenue undisclosed
VAST Data, earlier baseline	Approx. $100M ARR in late 2023	$9.1B, 2023	Shows how far its business has progressed
DDN, earlier baseline	Approx. $400M revenue in 2024	Before Blackstone investment	Established revenue base before the latest acceleration
WEKA, earlier baseline	Revenue doubled repeatedly through 2024	$750M, 2022	Demonstrates sustained earlier growth

VAST has moved into a different financial category from most venture-backed storage specialists. DDN remains the important exception because its business was already large before AI became the industry's main growth driver.

Is VAST Data's lead backed by real customers?

VAST Data's commercial lead is backed by major production customers, including a $1.17 billion agreement with CoreWeave and a large AI infrastructure commitment from Sharon AI.

The CoreWeave agreement, announced in November 2025, is particularly revealing. CoreWeave had already been working with VAST for years before signing the expanded contract, which made VAST a primary data foundation for its AI cloud. CoreWeave turned years of working with VAST into a bigger contract, and we separate who has real customers from who only promises in our AI infrastructure report.

VAST was able to turn an existing technical relationship into a much larger commercial commitment.

In June 2026, Sharon AI announced plans to deploy 600 petabytes of VAST infrastructure across Australia and Asia-Pacific. That is equivalent to 600,000 terabytes of data capacity. Sharon AI described the system as supporting infrastructure requirements comparable to those of approximately 100,000 GPUs, although the figure reflects sizing assumptions rather than a verified count of GPUs using the platform.

There is another interesting example outside conventional cloud computing.

Hyundai Motor Company and Kia made a strategic investment in VAST in October 2026 after several years of production use. Their autonomous mobility division, 42dot, had been using the technology for advanced driver-assistance research and autonomous driving workloads.

CoreWeave expanded a long-running cloud relationship, Sharon AI committed to a large regional infrastructure project, and Hyundai and Kia became investors after using the product themselves.

These are three different customer profiles making substantial commitments for different reasons. VAST isn't dependent on one AI laboratory or one particular type of GPU installation, and that's a valuable advantage.

Can WEKA actually beat VAST Data on AI storage performance?

WEKA can beat VAST Data in particular high-performance AI workloads, but there is no reliable public benchmark proving that WEKA is universally faster.

WEKA has spent years developing parallel file-storage technology for demanding computing environments. Its NeuralMesh platform is designed to handle large numbers of simultaneous data requests while keeping GPU clusters supplied with information.

The company has worked with Cohere, Midjourney, Stability AI, and several scientific computing organizations. These are credible examples because model development and research can place extreme demands on storage throughput.

WEKA has also pushed aggressively into inference infrastructure.

In March 2026, the company announced an integration with NVIDIA's STX reference architecture, built around BlueField-4 and Vera Rubin systems. WEKA estimated that its approach could deliver four to ten times more tokens per second for certain context-memory workloads.

Those figures come from WEKA's own performance projections. They do not establish what happens when customers compare WEKA and VAST under identical conditions.

VAST has its own high-performance architecture and NVIDIA integrations, while DDN has extensive experience with parallel storage in supercomputers.

A fair comparison would need to measure sustained GPU utilization, checkpoint recovery, throughput under heavy concurrency, and total infrastructure cost using the same workloads.

Until those results are publicly available, WEKA deserves recognition as a top technical competitor, but calling it the outright performance winner would be premature.

Why is DDN such a serious competitor in AI storage?

DDN is a major threat to newer AI storage startups because the company already knows how to operate storage infrastructure for enormous supercomputing installations.

Modern AI clusters share several requirements with traditional high-performance computing systems. Both need to feed large numbers of processors, move substantial datasets, and maintain reliable performance under sustained demand.

DDN built its business serving these environments long before ChatGPT made GPU infrastructure a mainstream investment theme.

Its relationship with NVIDIA stretches back years, and its systems have supported major supercomputing and AI installations, including infrastructure associated with xAI's Colossus project.

The company also received a $300 million investment from Blackstone in 2025 at a reported $5 billion valuation.

What's particularly interesting is the revenue trajectory. DDN's reported growth from $400 million to $500 million between 2024 and 2025 represents approximately 25%. Management's subsequent $1 billion projection would imply another 100% increase.

The second figure remains a forecast, but it shows the scale of business DDN is pursuing.

DDN could become the largest private AI-focused storage supplier by recognized annual revenue without ever being categorized as an AI-era startup.

An assessment limited to VAST, WEKA, and Hammerspace would therefore miss one of their most formidable competitors.

Is Hammerspace growing faster than VAST Data and WEKA?

Hammerspace has reported explosive bookings growth, but we still cannot establish whether its revenue is growing faster than VAST Data's or WEKA's.

In April 2026, Hammerspace announced that its bookings for the year had already reached nearly fourteen times the total recorded throughout 2025. Hammerspace booked nearly fourteen times last year's total; we compare who is ahead and who is falling behind in our report on AI infrastructure.

The comparison is unusual because the company exceeded a full year's bookings by such a large multiple before even completing the first four months.

There was also earlier momentum. Hammerspace reported tenfold revenue growth in 2024 alongside a 32% increase in customer count.

Together, those announcements suggest that the company has been winning much larger commercial opportunities over several years.

However, bookings represent contracted business, and the public announcement does not provide the absolute value needed to compare Hammerspace with its competitors.

A company moving from $5 million to $70 million in bookings would report the same fourteenfold increase as one moving from $100 million to $1.4 billion. Those are completely different competitive positions.

Hammerspace could be gaining ground very quickly. We just don't know how big that ground is yet.

Why would companies choose Hammerspace instead of VAST Data or WEKA?

Companies choose Hammerspace when they need to make data stored across different systems accessible to AI applications without undertaking a huge migration project.

Imagine a company with important information spread across several data centers, cloud providers, and research facilities. Moving everything into one new storage platform could take months, cost millions, and create security or regulatory problems.

Hammerspace tackles that problem through a global namespace and data orchestration technology. Applications can access files across different locations while the underlying storage remains distributed.

The company has worked with organizations including Meta and Los Alamos National Laboratory. In April 2025, it secured $100 million in strategic growth capital from investors including Altimeter Capital and ARK Invest.

Its approach has become increasingly relevant as enterprises try to connect AI systems to existing corporate information.

But VAST and WEKA are not standing still. Both support distributed data environments, and VAST has developed its own global namespace and data services.

The distinction is in the starting point. VAST often competes to become an organization's primary data platform. Hammerspace can win customers that prefer to keep their existing storage infrastructure and coordinate access across it.

For enterprises dealing with fragmented information, that can be a more practical proposition than replacing hardware. And in many cases, avoiding a huge migration is the deciding factor.

Are AI cloud companies choosing VAST Data over WEKA?

VAST Data currently has stronger publicly disclosed evidence of very large AI cloud contracts, although WEKA is also winning customers building and operating GPU infrastructure.

CoreWeave's billion-dollar agreement gives VAST a particularly strong reference in the AI cloud market. Its expanded relationship with Sharon AI adds a second example involving large, long-term infrastructure commitments.

WEKA has taken a somewhat different route.

In July 2026, the company announced a partnership with Andromeda to provide NeuralMesh as a storage layer for managed GPU clusters. Andromeda works across more than 50 compute providers worldwide.

That does not mean WEKA has been deployed across all 50 providers, but the partnership creates a way to reach customers operating in different clouds without requiring a separate large installation for each one.

WEKA has also pursued regional infrastructure partnerships, including Siam AI in Thailand and collaborations connected to Singapore's AI ecosystem.

VAST appears stronger in the largest publicly documented commercial commitments, while WEKA has credible adoption across AI-native businesses and distributed GPU environments.

What we cannot establish is their respective share of all AI cloud storage spending. Neither company publishes enough consistent contract data to calculate that accurately.

Does NVIDIA prefer VAST Data, WEKA or DDN?

NVIDIA is supporting several competing AI storage architectures rather than choosing one exclusive winner.

The chipmaker participated in financing rounds for both VAST Data and WEKA. It also maintains a long-standing engineering relationship with DDN, while Hammerspace is involved in NVIDIA's broader accelerated computing ecosystem. Nvidia joined financing rounds at two of these companies, and we follow who is getting funded, and by whom, in our report covering AI infrastructure.

The partnerships make sense when we examine the different problems these companies address.

VAST is integrating storage, databases, and data services more closely with NVIDIA's accelerated computing architecture. WEKA is developing persistent context-memory capabilities and high-performance data access. DDN contributes experience from large GPU and supercomputing installations.

NVIDIA's customers have different requirements. An AI laboratory training foundation models may want maximum sustained throughput. A business operating AI agents needs efficient access to stored context. A government building sovereign AI infrastructure may prioritize control over where data resides.

Supporting multiple suppliers gives NVIDIA more ways to sell complete AI systems.

For the startups, inclusion in NVIDIA reference architectures is valuable because it can make enterprise procurement and deployment easier. Still, a certified integration says little about how much business a company will ultimately win.

We shouldn't interpret NVIDIA's partnerships as a hidden ranking of storage performance.

Is AI inference changing who will win the storage market?

AI inference is giving VAST Data and WEKA new opportunities, especially as AI agents require much more persistent context and repeated access to information.

Training large models requires substantial storage throughput, but inference introduces its own bottlenecks.

A simple chatbot interaction might be handled almost entirely through GPU memory and ordinary retrieval systems. Longer-running AI agents can generate extensive conversation histories, cached attention states, intermediate results, and tool outputs that need to remain accessible across many interactions.

Keeping all of that information in expensive GPU memory becomes increasingly difficult as the number of concurrent users grows.

This is why WEKA has been developing its Augmented Memory Grid, which extends usable context memory beyond GPU memory. Its announced NVIDIA STX integration targets precisely these workloads.

VAST is moving in a similar direction through its BlueField-4-based inference architecture, which aims to make memory and data services available more efficiently across accelerated systems.

Hammerspace is taking another approach by helping inference applications reach data already distributed throughout an enterprise.

The leading storage challengers are all moving beyond their original role of feeding training datasets to GPUs.

The opportunity is attractive, but the size of the new business remains uncertain. Vendor performance estimates and architecture demonstrations have arrived faster than independently verified revenue figures for these products.

Is VAST Data becoming more than a storage company?

VAST Data is increasingly trying to become a complete AI data infrastructure platform, which helps explain why investors value it so much more highly than traditional storage businesses.

Its original technology focused on unified, high-performance storage. The product portfolio now includes the VAST DataBase, DataEngine, and services that help applications process and retrieve information.

In February 2026, VAST expanded its work with NVIDIA around a more integrated AI data stack, combining data processing, analytics, vector search, and retrieval capabilities.

A partnership with Cloudera in July connected VAST infrastructure with established enterprise data-management software.

In September, the company introduced DataEnclave, designed to let organizations bring AI models into controlled environments containing sensitive corporate information. Its announced ecosystem included Cohere, CrowdStrike, and Deepgram.

Across these launches, VAST is increasingly competing for budgets that would traditionally have gone to databases, analytics platforms, and AI application infrastructure.

That could make the company much larger than a specialist storage vendor.

The risk is that each expansion brings new competitors. Selling storage to GPU operators is different from persuading enterprises to replace existing database and data-processing tools.

VAST has already proved that it can win large storage contracts. Whether its newer services generate similarly strong revenue is still an open question.

Is VAST Data actually worth $30 billion today?

VAST Data's $30 billion valuation is expensive relative to its disclosed recurring commitments, although its growth and positive cash flow make the premium more credible than it would be for an unprofitable startup.

Using $500 million as the reference point for committed annual recurring revenue produces a valuation ratio of 60 times. Because VAST reports more than $500 million, its actual ratio to that metric is somewhat lower.

There are two reasons to treat that calculation carefully.

First, committed ARR includes contracted recurring business and can exceed revenue currently being recognized. Second, VAST's valuation increasingly reflects expectations about its broader AI data platform, rather than storage sales alone.

The increase from $9.1 billion in 2023 to $30 billion in 2026 represents approximately 230% appreciation.

Investors are therefore expecting continued rapid growth, additional large customers, and successful expansion into inference and enterprise data services.

Positive free cash flow makes that story more convincing, since the company has reported the ability to generate cash while expanding.

But the valuation leaves limited room for disappointment. If new products fail to generate meaningful sales or spending on AI infrastructure slows substantially, VAST could remain a successful business while struggling to justify its latest price.

We would rank VAST first operationally without automatically calling it the best investment in the sector. At $30 billion, a lot of future success is already in the price.

Could cheaper storage from Dell, Pure Storage or cloud providers hurt these startups?

Dell, Pure Storage, NetApp, and major cloud providers could put real pressure on AI storage startups as more AI workloads become standard enterprise applications.

Large technology suppliers already have extensive customer relationships, support teams, and purchasing agreements. Many businesses would prefer to buy another storage product from an existing supplier rather than introduce a new infrastructure company. Established suppliers already hold the customers and support teams; we look at how far the challengers are behind in our AI infrastructure report.

Price also matters, although the cheapest capacity isn't always the cheapest way to run AI workloads.

For a GPU cluster, slow storage can waste computing capacity. A system that costs more per terabyte may still be economically attractive if it reduces GPU idle time, speeds up checkpoints, or removes unnecessary data movement.

The harder question is how often customers actually need the most expensive storage architecture.

Large training clusters and highly demanding inference systems can justify specialized infrastructure. Smaller enterprise AI applications may work perfectly well with conventional cloud storage and caching.

That creates a risk for vendors whose products depend mainly on raw performance advantages.

VAST has responded partly through enterprise distribution partnerships, including Cisco. WEKA offers its software across different hardware and cloud environments, while Hammerspace can complement existing storage rather than replacing it.

The winner will increasingly depend on measurable customer economics: how much it costs to train a model, serve requests, and manage data over several years.

Public comparisons of those costs remain too limited to declare a universal winner.

Which AI storage startup is best for each type of customer?

VAST Data is currently the strongest all-round choice for large AI data infrastructure, while WEKA, Hammerspace, and DDN can be better suited to particular technical environments.

For a company building an entirely new AI cloud, VAST has a compelling record of large deployments and the ability to combine multiple data services on one platform.

WEKA deserves particular attention when GPU performance, parallel file access, and efficient context-memory handling are the main requirements.

Hammerspace becomes attractive when valuable data is already spread across storage systems that an organization cannot easily replace.

DDN is especially relevant for customers operating supercomputing environments, scientific workloads, and very large GPU installations where its experience and support infrastructure are useful.

We should also distinguish established strengths from newer product ambitions. For example, VAST has much stronger evidence in large-scale storage deployments than in replacing enterprise databases. WEKA's inference innovations are promising, but published vendor estimates are insufficient to establish a cost advantage across all applications.

Customer requirement	Strongest contender	Reason	Main question to test
Building a large AI cloud	VAST Data	Major contracts with AI infrastructure providers	Full deployment economics
Running intensive AI training	WEKA / DDN	Experience in demanding parallel workloads	Performance under the customer's dataset
Consolidating AI data services	VAST Data	Broad unified platform	Maturity of newer services
Connecting existing storage systems	Hammerspace	Global namespace and data orchestration	Performance across mixed infrastructure
Supporting scientific supercomputing	DDN	Long HPC operating history	Integration with newer AI services
Serving long-context AI agents	WEKA / VAST Data	Context-memory architectures	Verified cost per token at scale
Building sovereign AI infrastructure	VAST Data / WEKA	Regional cloud and AI partnerships	Local deployment, control, and support
Avoiding major storage migration	Hammerspace	Works with existing data environments	Compatibility and implementation cost

These are starting points for customer evaluations rather than universal technical rankings. A company's existing infrastructure and actual workload can change the result considerably.

What could make WEKA or Hammerspace overtake VAST Data?

WEKA or Hammerspace could challenge VAST Data's lead by turning their technical strengths into much larger recurring revenue and long-term customer contracts.

WEKA has the clearer route to becoming a direct rival in high-performance AI infrastructure. It already has an established product, major AI customers, and deep NVIDIA integrations.

What we need to see next is a more recent ARR disclosure showing how far it has grown beyond its 2024 milestone. A series of large contracts comparable to those signed by VAST would be particularly convincing.

Hammerspace has a different opportunity. Enterprises often have years of data scattered across incompatible environments, and that information becomes more valuable as AI systems move into production.

If Hammerspace can consistently convert its recent surge in bookings into recurring revenue, its market position could change quickly.

DDN is already operating at substantial scale. Delivering its projected annual revenue while extending its presence in large GPU environments would make it an even stronger competitor.

VAST faces its own test. The company has expanded well beyond storage, and investors now expect those additional capabilities to generate meaningful business.

There is also a broader industry risk. Large AI infrastructure projects rely on funding, electricity, data-center construction, and future customer demand. Recent concerns about the financing of heavily leveraged AI cloud providers show that announced capacity and long-term contracts do not eliminate execution risk.

A change in the ranking will become convincing when competitors demonstrate sustained revenue growth, substantial repeat contracts, and strong financial performance, rather than simply announcing another impressive technical benchmark.

So which AI storage startup is ahead today?

VAST Data is currently the clear overall leader among AI storage startups, with WEKA second and Hammerspace third. DDN belongs alongside the leading competitors, but its much longer history puts it in a different category.

VAST has built the strongest combination of contracted business, production customer relationships, partnerships, and positive financial performance.

Its lead has also become more convincing as the customer base has expanded beyond AI cloud providers. Automotive AI, sovereign infrastructure, enterprise data platforms, and inference services now form part of its broader strategy.

WEKA remains the closest direct technical challenger, especially for demanding GPU workloads and newer context-memory systems. Hammerspace has an appealing position in enterprise data orchestration, where replacing existing infrastructure is often impractical.

DDN deserves particular attention because its established revenue base and supercomputing experience give it advantages that newer companies cannot easily reproduce.

The important uncertainty now concerns the next phase of AI infrastructure. Training workloads helped create the market for specialized AI storage, but inference and long-running AI agents could change which architectures customers value most.

VAST has already proved that it can build a very large business around AI data infrastructure. WEKA and Hammerspace still have opportunities to lead particular parts of the market, especially where technical requirements differ from traditional storage.

For now, though, the evidence points clearly to VAST. It has the strongest commercial position, the broadest ambitions, and the clearest path toward becoming a major data infrastructure platform for AI.