The Hidden Data Layer Beneath xAI, Palantir, OpenAI-Linked Infrastructure and the U.S. Government

An anonymous former employee of a major AI infrastructure company recently made an extraordinary claim: the company they had worked for sat underneath systems used by some of the most powerful technology companies and governments in the world.

The source, whom we will call Bob, named Palantir, xAI, ChatGPT, Disney and NASA. Bob described an Israeli-founded data infrastructure company whose CEO had joked about wanting to “take over the world,” and said the concentration of sensitive data and artificial-intelligence infrastructure had become frightening enough that they quit.

We did not assume Bob was telling the truth.

Instead, we tried to identify the company from the clues and independently reconstruct its relationships.

They overwhelmingly point to VAST Data.

What we found is both less sensational and more consequential than the original message.

There is currently no evidence that VAST secretly possesses all of these organizations’ data, that sensitive U.S. government information is being transferred to Israel, or that Israeli personnel have unauthorized access to American systems.

But the documented reality deserves scrutiny on its own.

VAST Data has become a largely invisible infrastructure layer beneath some of the largest AI systems in existence. It powers xAI’s Colossus supercomputer, serves as CoreWeave’s “primary data foundation,” publicly acknowledges a relationship with Palantir, supplies technology used by the U.S. Department of Defense and other federal agencies, supports a massive veterans’ genomic research program, and is building infrastructure meant to coordinate data, memory, identity and activity for potentially billions of autonomous AI agents.

VAST says its technology already underlies approximately 2 million to 2.5 million GPUs worldwide. At NVIDIA GTC DC in October 2025, co-founder Jeff Denworth said individual VAST systems can serve data centers containing 100,000 to 300,000 GPUs.

In April 2026, investors valued the company at $30 billion.

The important question, therefore, is not whether an obscure company is secretly “hosting everyone’s data.”

It is:

How did one relatively unknown infrastructure company become embedded beneath so much of the emerging AI economy—and what does that concentration mean as AI systems become increasingly autonomous?

The anonymous message that started the investigation

Bob described working for what they called “the most powerful AI data infrastructure company in the world.”

Among the claims:

• the company provided infrastructure involving Palantir;
• it supported xAI, NASA, Disney and ChatGPT;
• its founder was an Israeli computer scientist;
• the company had major operations in Tel Aviv;
• the founder had joked at a company gathering about taking over the world, invoking Pinky and the Brain;
• the company had extensive connections to governments and militaries.

Several more personal claims—including an allegation that the founder repeatedly emailed employees saying he could not feel emotion—cannot presently be independently verified.

Those claims should therefore remain allegations unless original messages or additional corroborating witnesses emerge.

But the corporate fingerprints were unusually specific.

They led to VAST Data.

And once we began checking the public record, most of the important infrastructure clues had real-world counterparts.

What is VAST Data?

VAST began as a high-performance data-storage company but increasingly rejects the description of itself as merely a storage vendor.

Its current product is called the VAST AI Operating System.

The company says the platform combines storage, database services, global data access, event processing, vector search and compute functionality into a common architecture intended to support AI systems from raw-data ingestion through inference and autonomous action.

That matters because AI infrastructure has several layers.

GPUs perform enormous amounts of computation. Models provide the intelligence people interact with. But those systems also require infrastructure capable of continuously feeding GPUs datasets, storing checkpoints, retrieving information, maintaining metadata and moving information between training and inference systems.

That is the layer VAST increasingly occupies.

Denworth described its footprint at NVIDIA’s 2025 GTC DC conference this way:

“VAST is the data platform for somewhere in the order of 2 to 2.5 million GPUs deployed around the world today.”

He also described single data centers containing 100,000 or even 300,000 GPUs sharing a VAST system.

Those figures are VAST’s own claims, not independently audited measurements. But their scale helps explain why a company unfamiliar to most consumers can nonetheless occupy an important position in the AI supply chain.

xAI’s Colossus runs its data workloads on VAST

The xAI connection is not speculation.

VAST founder and CEO Renen Hallak publicly described VAST as the “data platform behind xAI’s Colossus cluster” and said xAI runs data and workloads on the VAST Data Platform.

VAST co-founder Jeff Denworth went further, writing publicly that xAI:

“trains on VAST Data, checkpoints on VAST, and protects data with VAST.”

That is considerably more specific than a generic customer relationship.

Training frontier AI models requires continuously feeding massive datasets into enormous GPU clusters. Model checkpoints—the saved states that allow training to resume or preserve intermediate model versions—can themselves be enormous.

For xAI’s Colossus, VAST publicly says its technology participates directly in those processes.

Bob’s simplified description that VAST “hosts xAI’s data” therefore captures something real, although “provides critical data infrastructure for xAI’s training system” is the more technically accurate description.

Disney is real too—but specifically through Pixar

Bob also named Disney.

That connection is straightforward.

Pixar Animation Studios, owned by Disney, selected VAST to support the data infrastructure behind its production pipeline.

For Elemental, Pixar moved approximately 7.3 petabytes of data into a VAST datastore cluster, according to VAST and Pixar. The system was designed to handle rendering workloads involving hundreds of thousands of processors and multiple simultaneous film productions.

Pixar’s infrastructure team also discussed the possibility that the datasets stored through the platform could support future AI and deep-learning applications.

So “Disney data” is directionally correct but imprecise.

The publicly established relationship is with Pixar’s production infrastructure.

NASA and multiple federal agencies also appear in VAST’s public customer record

VAST Federal has publicly identified federal customers or supported organizations including:

• the Department of Defense;
• Department of Energy;
• National Institutes of Health;
• National Oceanic and Atmospheric Administration;
• NASA;
• Department of Veterans Affairs.

The NASA relationship remains less transparent than the xAI or Pixar deployments. We have not found a sufficiently detailed public record establishing exactly which NASA datasets, systems or facilities use VAST.

That is a worthwhile public-record target.

But the broader federal footprint is indisputable.

The Pentagon bought more than $10 million of VAST systems years ago

In August 2021, VAST Federal announced that Department of Defense orders for its storage technology had exceeded $10 million.

VAST said the deployments included demanding national-security workloads involving synthetic-aperture radar imagery and network packet capture, with AI and analytics used to process information quickly.

That federal relationship has since matured.

In April 2025, the Defense Information Systems Agency’s Joint Interoperability Test Command formally certified the VAST Federal Data Platform, software release 5.2, as a Data Storage Controller for joint Department of Defense use.

The government certification says the system met the applicable critical interoperability requirements with “no conditions.”

This is important evidence because it does not come from VAST’s marketing department.

It comes from the Department of Defense itself.

VAST also supports one of the world’s largest veterans’ genomic research programs

One federal deployment illustrates just how sensitive the systems can become.

VAST was selected to support infrastructure for the Department of Veterans Affairs’ Million Veteran Program, one of the world’s largest genomic and health-research initiatives.

The program combines genetic information with health records, lifestyle information and military-exposure data to study disease and health outcomes among veterans.

VAST said the implementation involved more than 30 petabytes of genetic information.

That does not mean VAST employees can freely inspect veterans’ genomes.

The distinction is crucial.

It means VAST software is trusted to operate within an environment containing extraordinarily sensitive data.

The security architecture governing access to that environment is therefore not a minor technical detail. It is central to understanding the relationship.

VAST has explicitly expanded into the intelligence and defense ecosystem

VAST Federal’s national-security ambitions have never been hidden.

In 2021, the company appointed former CIA Chief Technology Officer Gus Hunt to its board.

Hunt served as CIA CTO from 2008 through 2013 and previously directed application services for the agency. VAST said his role would include advising the company’s federal expansion.

Again, employing a former intelligence official does not establish misconduct.

But it demonstrates how deliberately VAST positioned itself as infrastructure for national-security customers.

The company has since moved considerably beyond high-speed storage.

In 2025, VAST Federal and defense contractor Leidos announced an “agentic cybersecurity” collaboration using NVIDIA technology and VAST’s AI Operating System.

The proposed systems are designed to retain years of security telemetry and allow AI agents to correlate incidents, construct timelines and potentially recommend or execute actions, including rerouting traffic or closing network ports.

That progression is important:

storage → data management → analytics → AI orchestration → autonomous action

VAST’s strategic significance grows considerably if its software becomes not merely where information resides, but one of the systems through which AI agents retrieve information and act upon it.

The Palantir relationship is real

One of Bob’s most provocative claims involved Palantir.

There is public evidence for a relationship.

In November 2024, a user on VAST’s own Cosmos community asked whether VAST worked with Palantir.

An account identified simply as “jeff” responded:

“Yes, we do have a relationship with Palantir.”

The response described the companies as operating at complementary layers:

VAST at the data-infrastructure layer and Palantir at the data-processing layer.

It added that the relationship was in its early days and predicted further developments.

That answer raises more questions than it resolves.

Palantir’s Foundry and Artificial Intelligence Platform can operate on extremely sensitive commercial and government datasets.

If VAST operates underneath Palantir deployments, the architecture could potentially resemble:

customer data → VAST infrastructure → Palantir processing/application layer

But we have not established that this topology exists in any particular government system.

It would therefore be irresponsible to imply that VAST automatically holds or has access to every dataset processed through Palantir.

What is established is narrower and still significant:

VAST and Palantir publicly acknowledged that they were working together, with VAST describing itself as the underlying data-infrastructure layer.

What became of that relationship after 2024 remains largely unexplained publicly.

What about ChatGPT and OpenAI?

This is where Bob’s claim requires the most careful wording.

We have not found public evidence proving that VAST directly stores ChatGPT conversations or that OpenAI is itself a disclosed VAST customer.

But there is now a documented infrastructure chain connecting the companies.

In November 2025, CoreWeave signed a $1.17 billion agreement with VAST, describing VAST as the “primary data foundation” for CoreWeave’s AI cloud.

VAST said CoreWeave could deploy its platform in any data center for any customer.

CoreWeave, meanwhile, has one of the world’s largest commercial infrastructure relationships with OpenAI.

Its initial March 2025 OpenAI agreement was worth up to $11.9 billion. Further expansions increased CoreWeave’s total OpenAI commitments to approximately $22.4 billion by September 2025. CoreWeave says the infrastructure supports OpenAI’s model training and deployment.

The documented chain therefore looks like this:

VAST → CoreWeave → OpenAI

But that is not the same thing as:

VAST → ChatGPT user conversations

Those are radically different claims.

CoreWeave operates enormous infrastructure environments, and the fact that VAST is CoreWeave’s primary data foundation makes it entirely plausible that some OpenAI-related workloads could interact with VAST systems.

It does not prove which workloads do.

Until a contract, architecture diagram, customer record, support ticket or other technical documentation closes that final gap, the responsible conclusion is:

VAST is directly embedded in a major infrastructure provider serving OpenAI, but we cannot establish that VAST stores ChatGPT user data.

That distinction is not semantic nitpicking.

It is the difference between evidence and assumption.

“VAST hosts everyone’s data” is technically misleading

This is also where the viral version of the story can go badly wrong.

VAST does not necessarily operate some centralized warehouse where all customer information is copied into company-controlled servers.

The VAST AI Operating System is software-defined.

VAST says customers can deploy it on infrastructure from their chosen hardware vendors and cloud platforms.

That means a customer’s data may remain in the customer’s own data center while VAST software manages the storage and data services.

Several concepts must therefore be separated:

Infrastructure provider does not automatically mean data owner.

Data-management software does not automatically mean remote access.

VAST software operating on a system does not automatically mean VAST employees can inspect everything stored there.

And an Israeli parent or corporate entity does not establish that customer information is transferred to Israel.

So far, we have found no evidence supporting the claim that VAST secretly transfers U.S. government, OpenAI, Palantir or xAI production data to Israel.

But we did find another data pathway that deserves much closer scrutiny.

VAST systems can transmit diagnostic information back to VAST

VAST’s current documentation describes a service called Uplink, connected to its Callhome and support systems.

When enabled, a VAST cluster can report configuration information, metrics, health information and diagnostic or log bundles into VAST’s support infrastructure.

VAST says:

• communications are initiated by the customer’s cluster;
• communication between Uplink and the cluster is one-way;
• VAST personnel cannot directly access a customer cluster unless the customer explicitly grants that access;
• cluster configuration information and metrics may be stored in cloud databases;
• support and log bundles may be stored in Google Cloud Storage;
• authenticated VAST support personnel can access certain reported information through internal systems.

VAST separately says support bundles can contain logs, metrics and diagnostic information and are retained for at least 12 months, unless deleted earlier at customer request.

The company says geographic data-residency restrictions can also be configured.

Its August 2026 documentation further says customers creating a support bundle may choose to obfuscate sensitive data before sending it.

These are important safeguards.

They also reveal exactly where the next investigation needs to go.

For sensitive deployments involving the Pentagon, veterans’ genetic information, NASA, xAI, CoreWeave, Palantir-linked systems or foreign sovereign infrastructure:

Is Uplink enabled?

What information is actually collected?

Where does it physically reside?

Which VAST personnel can access it?

Are support personnel restricted by citizenship or geography?

Do federal deployments use separate support infrastructure?

Is obfuscation mandatory for sensitive systems or merely optional?

Are classified or air-gapped deployments configured differently?

The documentation tells us what the technology can do.

It does not tell us how every sensitive customer configures it.

That is the important unanswered question.

VAST’s Israeli roots are real—but the corporate story is more complicated than “headquartered in Tel Aviv”

Bob described the company as headquartered in Tel Aviv.

Today, VAST publicly presents itself as a remote-first company with major American operations and routinely issues corporate announcements from New York.

But the Israeli corporate structure is more substantial than a simple foreign engineering office.

An active Israeli private corporation named VAST DATA LTD, company number 515350155, was incorporated on December 2, 2015 and lists an address at 97 Rokach Boulevard, Tel Aviv. Israeli corporate-data services continue to identify the company as active.

U.S. intellectual-property records also identify that Israeli entity.

A U.S. Patent and Trademark Office trademark record lists:

Vast Data Ltd. — Israel — 97 Rokach Blvd., Tel Aviv.

A separate patent assignment filed with the USPTO transfers intellectual property to VAST DATA LTD, described as having its principal place of business at the same Tel Aviv address.

This does not establish foreign government control.

It does mean the corporate structure should be described accurately.

VAST is simultaneously deeply embedded in the U.S. technology and federal markets and legally rooted through a significant Israeli corporate entity that owns intellectual property associated with the platform.

For sensitive government technology, corporate control, jurisdiction, support access and contractual restrictions are legitimate questions regardless of which foreign country is involved.

Nationality itself is not evidence of wrongdoing.

VAST is also becoming infrastructure for “sovereign AI”

The concentration extends beyond the United States.

In September 2025, VAST announced that its software would power a major Israeli sovereign AI cloud operated with Shonfeld Data Services, involving thousands of NVIDIA Blackwell GPUs and dozens of petabytes of data infrastructure.

Two months later, VAST announced a collaboration with Lablup supporting South Korea’s government-backed Sovereign AI Project, including model-training infrastructure running on SK Telecom’s sovereign AI cluster.

And in June 2026, Australian AI-cloud company Sharon AI announced a planned 600-petabyte VAST deployment, positioning VAST as the foundational data layer for sovereign AI infrastructure serving government, research and enterprise customers across Australia and the Asia-Pacific region.

The interesting contradiction is obvious.

Nations increasingly talk about AI infrastructure in terms of sovereignty—keeping strategic information, computing capacity and intellectual property under national control.

Yet the software layer underneath several supposedly sovereign systems can itself come from the same multinational infrastructure vendor.

That is not necessarily a security failure.

But it illustrates just how much geopolitical importance can accumulate inside companies almost entirely invisible to the public.

VAST is no longer trying to be merely a storage company

This may ultimately be the most important part of the story.

VAST’s ambitions have moved dramatically upward through the technology stack.

Its platform has evolved from storage into a system encompassing:

• file and object storage;
• block storage;
• databases;
• streaming data;
• vector search;
• serverless functions;
• AI pipelines;
• global namespaces;
• policy and identity controls;
• compute orchestration.

VAST now calls the result an AI Operating System.

In February 2026, the company introduced Polaris, described as a global control plane capable of provisioning and orchestrating distributed VAST infrastructure across public clouds, private data centers and other AI environments.

Customers can operate VAST infrastructure themselves, and the existence of a control plane does not mean VAST centrally commands every customer’s systems.

But the strategic direction is unmistakable.

The company wants its technology to become an increasingly unified layer between physical AI infrastructure and the intelligent systems running above it.

The real ambition: memory and coordination for billions of AI agents

This brings us back to Bob’s strangest anecdote.

Bob claimed CEO Renen Hallak jokingly told approximately 1,000 employees or attendees that he wanted to “take over the world” like Pinky and the Brain.

We have not verified that specific event.

Curiously, however, essentially the same joke appeared publicly years earlier from VAST co-founder Jeff Denworth, suggesting it may indeed have circulated as company humor.

But the joke is almost a distraction.

Hallak’s publicly documented serious vision is considerably more interesting.

At VAST Forward 2026, he argued that the next generation of computing infrastructure will need to coordinate not merely files or databases, but the:

data, memory, identity, communication and activity of billions of AI agents.

VAST’s own account of the presentation describes its global data architecture as beginning to resemble a “planetary memory layer.”

The article explaining Hallak’s vision says billions of agents could communicate with one another, access sensitive information, generate models and act in the physical world.

Someone, it argues, will need to:

• track their actions;
• enforce policy;
• manage identity;
• maintain memory;
• coordinate their activity across data centers, clouds, edge devices and robots.

Hallak’s answer is that this responsibility belongs to the AI operating system.

And VAST intends to build it.

This is the central issue.

A storage vendor is useful.

An infrastructure layer responsible for the memory, data access, permissions and orchestration of enormous populations of autonomous agents becomes something qualitatively different.

It becomes part of the governance architecture of artificial intelligence itself.

So should VAST Data’s reach “terrify” people?

The evidence supports concern and scrutiny.

It does not support panic.

Bob’s message bundles together several ideas that need separating.

Verified

VAST provides critical data infrastructure to xAI.

VAST is CoreWeave’s primary data foundation.

CoreWeave has approximately $22.4 billion in OpenAI infrastructure commitments.

VAST publicly acknowledges a Palantir relationship.

Pixar uses VAST for petabyte-scale production data.

VAST technology is used across multiple U.S. government organizations.

The Pentagon has purchased VAST systems.

A DoD authority has certified the VAST Federal Data Platform.

VAST supports infrastructure associated with a massive veterans’ genomic research program.

VAST is increasingly building orchestration and agentic-AI infrastructure rather than merely storage.

VAST’s CEO publicly envisions infrastructure coordinating billions of AI agents.

Plausible but not established

Some OpenAI workloads running through CoreWeave may use VAST infrastructure.

VAST may sit underneath Palantir at mutual customer deployments.

Some sensitive customers may transmit diagnostic information into VAST support infrastructure.

All of those are reasonable investigative hypotheses.

None should presently be published as established fact.

Unsupported by the evidence we found

VAST secretly possesses everyone’s data.

VAST stores all ChatGPT conversations.

Sensitive U.S. government data is being secretly sent to Israel.

Israeli employees can freely access U.S. defense or veterans’ genomic information.

VAST has an improper relationship with Israeli intelligence.

We found no evidence establishing those claims.

The whistleblower now needs to prove what the public record cannot

This creates an unusual situation.

Bob’s message initially sounded almost impossibly sweeping.

Yet many of its underlying corporate clues survived scrutiny.

VAST really does power xAI’s data infrastructure.

It really does work with Palantir.

It really does serve sensitive government environments.

It really does have significant Israeli corporate roots.

It really is deeply connected to CoreWeave, which in turn supplies enormous amounts of infrastructure to OpenAI.

And its leadership really does describe a future in which VAST-like infrastructure manages the memory, identity and coordination of enormous populations of AI agents.

That makes Bob more interesting as a source.

But it also raises the standard for what Bob must provide next.

The public record has taken us surprisingly far.

A genuine whistleblower should be able to take us somewhere the public record cannot.

The most valuable evidence would be documentation answering questions such as:

Was OpenAI itself an identifiable VAST workload or account?

What exactly was the architecture of the Palantir relationship?

Which government systems had Callhome or Uplink enabled?

Could employees outside the United States access federal support bundles or telemetry?

What geographic or citizenship restrictions applied to those employees?

Were customers promised restrictions internally that differed from actual practice?

What was the specific event that caused Bob to resign?

And above all:

What did Bob personally witness that the public does not already know?

That is where a whistleblower investigation becomes something more than an unusually good tip.

The bigger story is infrastructure concentration

AI debates are usually about the companies consumers can see.

OpenAI.

Google.

Anthropic.

xAI.

Microsoft.

Meta.

Palantir.

NVIDIA.

But those companies depend on another layer of infrastructure that receives much less scrutiny.

The companies controlling that layer can potentially become extraordinarily important without ever becoming household names.

VAST appears to be one of them.

At a $30 billion valuation, it is already among the world’s most valuable private infrastructure companies. It says its systems support millions of GPUs. Its technology sits beneath some of the world’s largest AI environments, enters national-security systems, supports enormous scientific datasets and increasingly extends from storage into databases, inference pipelines, agent orchestration and global control planes.

None of that means VAST is doing something improper.

It means the company deserves substantially more scrutiny than it receives.

Because if the future of AI really does involve billions of autonomous agents continuously reading information, building memories, talking to one another and taking actions, then the companies controlling the infrastructure underneath those agents will wield a form of power most discussions about AI governance barely acknowledge.

The question is no longer just:

Who owns the AI model?

It is also:

Who owns and operates the systems that give that intelligence its data, memory, permissions and ability to act?

VAST Data is betting heavily that the answer will include VAST Data.

That is worth understanding before the infrastructure becomes so ubiquitous that nobody remembers to ask.

References and Further Reading

VAST Data scale, corporate strategy and architecture

• VAST Data Valued at $30 Billion as AI Drives a New Infrastructure Stack — VAST’s April 2026 Series F announcement, including its $30 billion valuation and description of its AI-infrastructure footprint.

• Building the AI Data Factory — NVIDIA GTC DC — NVIDIA-hosted presentation in which VAST co-founder Jeff Denworth says VAST serves roughly 2 million to 2.5 million GPUs and discusses extremely large deployments.

• The VAST AI Operating System White Paper — Detailed technical explanation of VAST’s architecture and evolution from storage into a broader AI operating system.

• VAST FWD 2026: Renen Hallak on Building the New Infrastructure of Intelligence — VAST’s account of Hallak’s vision for infrastructure coordinating the memory, identity and activity of billions of AI agents.

• VAST Introduces Polaris — Announcement describing VAST’s global control plane for distributed AI infrastructure.

xAI, CoreWeave and OpenAI

• Renen Hallak on VAST and xAI’s Colossus — Hallak’s public description of VAST as the data platform behind xAI’s Colossus cluster.

• VAST and CoreWeave’s $1.17 Billion Infrastructure Agreement — Establishes VAST as CoreWeave’s primary data foundation.

• CoreWeave Expands OpenAI Agreement to Approximately $22.4 Billion — CoreWeave’s primary announcement documenting the scale of its OpenAI relationship.

• CoreWeave Announces Initial $11.9 Billion OpenAI Infrastructure Agreement — Documents CoreWeave’s role providing training and service infrastructure to OpenAI.

Palantir

• VAST Cosmos Community: Palantir Relationship — Public VAST community discussion in which a VAST representative says VAST and Palantir are working together, with VAST at the data-infrastructure layer.

Federal, defense and health infrastructure

• VAST Federal Announces More Than $10 Million in Department of Defense Orders — Describes early DoD deployments including radar and network-data workloads.

• DISA Joint Interoperability Certification of the VAST Federal Data Platform — Primary Defense Department certification document.

• VAST Data Selected for the Million Veteran Program — Describes VAST infrastructure supporting VA genomic research.

• Former CIA CTO Gus Hunt Joins VAST Federal — Documents VAST Federal’s national-security expansion and Hunt’s board appointment.

• VAST Federal and Leidos Introduce Agentic Cybersecurity — Shows VAST’s movement from storage into AI-driven cybersecurity orchestration.

Support, remote telemetry and data residency

• About VAST Uplink — Current VAST technical documentation describing cluster-initiated telemetry, cloud storage and restrictions on direct customer-cluster access.

• VAST Platform Support Bundles — Describes diagnostic collection, access controls, retention and geographic data-residency options.

• Sending a Support Bundle Through the VAST Web UI — August 2026 documentation explaining what diagnostic bundles collect and the option to obfuscate sensitive information.

Corporate structure and Israeli entity

• U.S. Patent and Trademark Office Record for VAST Data Ltd. — Identifies VAST Data Ltd. as an Israeli entity at 97 Rokach Boulevard in Tel Aviv.

• USPTO Patent Assignment to VAST Data Ltd. — Patent assignment naming VAST Data Ltd. and its Tel Aviv principal place of business.

• Israeli Corporate Record Summary for VAST Data Ltd. — Registry-derived corporate information identifying VAST Data Ltd., company number 515350155, incorporated in Israel in 2015.

Sovereign AI deployments

• VAST and SDS Build Israeli Sovereign AI Cloud — Documents VAST’s role in a major Israeli sovereign AI infrastructure project.

• VAST and Lablup Support Korea’s Sovereign AI Initiative — Documents VAST infrastructure in South Korea’s government-backed sovereign AI project.

• Sharon AI Expands to a 600-Petabyte VAST Deployment — Documents the planned VAST data foundation for sovereign AI infrastructure across Australia and Asia-Pacific.

Editorial currency note: VAST Data’s customer relationships, corporate structure, software capabilities, federal deployments and support architecture are evolving rapidly. This investigation reflects publicly available records reviewed through September 4, 2026. Claims attributed to VAST, its executives or partners are identified as such where independent documentation is not available. The article should be updated if VAST, OpenAI, CoreWeave, Palantir or relevant government agencies provide additional documentation or responses.