A unicorn valued at $210 billion is about to be born, attracting the attention of NVIDIA, Google, and xAI simultaneously.

According to a report by Zhidx on August 26, recently, CapitalG, the venture capital arm of Google's parent company Alphabet, and NVIDIA are in talks to invest in Israeli AI infrastructure provider VAST Data. The financing amount may reach billions of dollars, which could become the largest - scale financing in the history of Israeli technology companies. After the financing is completed, the valuation of this startup will soar to $30 billion (approximately RMB 214.8 billion).

What exactly is the background of this startup that Google and NVIDIA are vying to invest in?

Founded in 2016, VAST Data has become a favorite among many large - model enterprises. The core reason is that the traditional data storage architecture cannot meet the new requirements of large - model training and inference. VAST Data has launched a unified data platform for the AI era, integrating structured and unstructured data to enable more efficient and cost - effective AI data processing.

Many globally renowned enterprises, such as xAI, a large - model startup under Elon Musk, CoreWeave, which received a $3.96 billion investment from NVIDIA, Disney, a global leading animation company, and its subsidiary Pixar, Verizon, a US telecommunications giant, and Zoom, a video - conferencing platform, are all on the customer list of this startup.

Notably, it has signed long - term contracts of 5 to 7 years with many customers, resulting in an explosive increase in annual revenue. According to the public data on VAST Data's official website, for the year ending January 31, 2025, VAST Data's revenue increased 3.6 times year - on - year. This revenue growth rate even exceeded that of NVIDIA and OpenAI. In fiscal year 2025, NVIDIA's revenue increased 114% year - on - year. Previously, according to Bloomberg, OpenAI's revenue is expected to triple year - on - year to $12.7 billion (approximately RMB 90.9 billion) in 2025.

According to a media report citing an anonymous source familiar with the financial situation, Renen Hallak, the co - founder and CEO of VAST Data, once mentioned that the company has achieved positive free cash flow for five consecutive years. As of January 2025, the company's ARR (Annual Recurring Revenue) reached $200 million (approximately RMB 1.43 billion), and it is expected to grow to $600 million (approximately RMB 4.3 billion) next year.

Meanwhile, this startup also has a close relationship with the two "financial backers" that are reported to be involved in the new financing. VAST Data has integrated its software platform into Google Cloud. Jensen Huang, the founder and CEO of NVIDIA, has repeatedly praised VAST Data at international conferences such as the GTC conference and the Taipei International Computer Show COMPUTEX, calling it a key enabler for large - scale AI model deployment.

In the venture capital circle, this startup has long been in the spotlight of investors. It has previously raised funds in five rounds, with a total financing amount of $381 million (approximately RMB 2.73 billion). At that time, its valuation reached $9.1 billion (approximately RMB 65.2 billion). Dell and NVIDIA have made consecutive investments in multiple rounds.

▲VAST Data's financing situation

The AI - driven data is growing at an unprecedented scale, which poses higher requirements for data - processing infrastructure, presenting development opportunities for VAST Data, which builds AI infrastructure for data processing.

01. Holding $1 billion in orders, with xAI, CoreWeave, and Disney as customers

The four founders of VAST Data all have profound experience in the storage field.

CEO Renen Hallak, CTO Shachar Finblit, Vice President of Marketing Jeff Denworth, and CTO Alon Horev co - founded the company in 2016.

▲VAST Data's co - founders Jeff Denworth (left), Shachar Finblit (middle), and CEO Renen Hallak (second from right)

Hallak once served as the vice president of R & D in the XtremIO department of Dell EMC's all - flash enterprise - class storage array. He led the project from its launch to achieving over $1 billion in revenue. Finblit and Horev have both worked at companies such as IBM, and Denworth has over 20 years of technical experience in advanced computing and large - scale, scalable big data and cloud storage.

At that time, Hallak felt the challenges of data storage for large - scale AI analysis in XtremIO, but he had no room to display his skills in the Dell department. So he decided to leave and build a new architecture from scratch. Subsequently, he hit it off with the other three co - founders.

In a nutshell, VAST Data's business system unifies storage, database, and containerized computing engine services into a single, scalable VAST Data software platform. This platform is designed from the ground - up for AI and GPU - accelerated tools in modern data centers and clouds.

Specifically, it can enable real - time access to unstructured data such as emails, logs, PDF files, and multimedia content. By moving non - critical data to lower - cost flash storage and then using faster and more expensive flash, GPUs can quickly access large amounts of data during model training.

Thanks to the increasing demand related to AI, many large - model - related enterprises and leading companies in other sectors have extended olive branches to VAST Data. In addition to the popular xAI and CoreWeave in the large - model field mentioned at the beginning, there are also Lambda, a cloud - computing infrastructure company invested by NVIDIA, Core42, a subsidiary of the UAE AI company G42, as well as other enterprises in different sectors such as NASA, the US Department of Energy, Boston Children's Hospital, and the travel company Booking Holdings. All are customers of VAST Data.

▲Partial customer list of VAST Data

Different from many software companies that rely on short - term contracts, VAST Data signs long - term contracts of 5 to 7 years with customers, resulting in an extremely low customer churn rate and cumulative software bookings of over $1 billion (approximately RMB 71.6 billion).

In terms of financing, VAST Data has previously raised more than $381 million in total, with a valuation of $9 billion. Top investment institutions such as Tiger Global and Goldman Sachs, as well as leading companies such as NVIDIA and Dell, are among its investors. It is worth noting that CapitalG, an independent growth fund of Alphabet, is reported to be participating in this round of financing. The investment goal of this fund is profit rather than strategic investment, which also indicates to some extent investors' recognition of VAST Data's profitability.

So, what makes VAST Data's products stand out?

02. Built specifically for AI needs, with a self - developed distributed system architecture

The ability to process data efficiently at low cost has always been one of the key factors in AI development.

Traditional data storage relies on a hierarchical structure, using low - cost storage solutions for long - term data storage and high - end solutions for more frequently used data.

However, the difficulty in data management lies in that, under the traditional architecture, it has become increasingly difficult to handle the transmission of PB - or even EB - scale data across global data centers; the traditional data architecture is not designed to meet today's AI requirements for massive, diverse data sets and high - performance random I/O; current solutions are too costly, forcing enterprises to make trade - offs between performance, scale, elasticity, and cost when managing and activating data.

Therefore, it is necessary to build a data - processing architecture specifically designed for AI.

VAST Data's approach is to eliminate the hierarchical model of traditional storage, storing structured, semi - structured, and unstructured data in one place to accelerate data retrieval and reduce the cost of model training and inference. Its greatest advantage is that it can prevent the tens of thousands or even hundreds of thousands of GPUs deployed by xAI and CoreWeave from being idle while waiting for storage.

How do they achieve this?

The company has proposed the first distributed system architecture DASE (Disaggregated Shared - Everything). As a proprietary framework specifically designed for AI needs, it unifies the storage, computing, and database layers into a single, globally consistent system. Different from public - cloud providers that stack different tools, VAST's AI operating system eliminates performance compromises and supports real - time analysis, recursive computing, and seamless hybrid - cloud operations.

Specifically, the disaggregated feature of DASE allows for the separation of data storage and computing resources, enabling each component to be independently and flexibly scaled; the shared - everything concept means that data can be accessed across all storage nodes, providing a unified data view for all nodes.

The VAST Data Platform built on this architecture can support a cluster of 10,000 GPUs, with a processing capacity of up to TB per second.

The VAST Data Platform operating system consists of many components:

VAST DataSpace allows data access, transactions, and protection from the edge to the cloud across hundreds of locations, similar to the global resource manager of an operating system; VAST DataStore is a general - purpose storage platform, comparable to the file system in an operating system; VAST DataBase is responsible for indexing functions and can provide multiple functions for real - time querying and analyzing data. VAST DataEngine plays the role of a dynamic computing and execution layer, and VAST InsightEngine is its internal data refinement tool, which uses AI embedding models to transform raw unstructured data into contextual data and serves as a RAG (Retrieval - Augmented Generation) tool.

▲VAST Data AI operating system architecture

In the second half of this year, VAST Data will complete the last piece of the puzzle for the core services of its data - processing AI operating system, the AI Agent deployment and orchestration system VAST AgentEngine. This means that VAST Data's platform has integrated the full - process capabilities of receiving data, storing it in real - time, and providing data to agents seeking information.

Looking at specific cooperation customers, in February this year, xAI officially announced its super - computing cluster Colossus, which is equipped with more than 200,000 NVIDIA GPUs. The data platform behind it was built by VAST Data, reducing the total cost of ownership (TCO) of Colossus' AI workloads by 50%. In September 2023, VAST Data and CoreWeave announced a strategic cooperation. CoreWeave built a global NVIDIA - accelerated computing cloud based on VAST Data's platform, which can manage and protect the large amounts of data required for generative AI, high - performance computing (HPC), and visual effects (VFX) tasks.

VAST Data's system, built from scratch for AI, unifies storage, database, and virtualized computing engine services. This also shows that in the face of new development opportunities in the AI industry, VAST Data has expanded from its initial positioning as a storage company to a broader application space.

03. Deeply tied to Google and NVIDIA, new financing may signal an IPO

The two giants reported to be making investments this time have both established deep partnerships with VAST Data.

First, NVIDIA. In March this year, VAST Data obtained NVIDIA's certified storage qualification. Jensen Huang mentioned VAST Data in his keynote speeches at the GTC conference and the Taipei International Computer Show COMPUTEX. He believes that in the AI era, data is the raw material driving the industry. NVIDIA is working with global storage leaders to build a new generation of enterprise infrastructure, and enterprises need to deploy and scale AI agents in hybrid data centers. VAST Data is one of the enterprises cooperating with NVIDIA.

At the end of 2024, Jensen Huang and Hallak recorded a ten - minute podcast about the future of AI. Huang talked about the data flywheel for continuous model improvement that he mentioned at the VivaTech conference in Paris, France. The current shift from training to real - time inference in enterprise AI expansion presents an excellent opportunity for VAST Data. He is very proud of the previous cooperation with VAST Data.

▲Jensen Huang and Hallak recording an AI podcast

In September last year, VAST Data also collaborated with NVIDIA to build the real - time RAG tool InsightEngine, which can use NIM microservices for real - time data retrieval and has achieved enterprise - level applications in fields such as financial transactions, autonomous driving, and logistics.

Second, Google. In April this year, the VAST Data platform was fully integrated into Google Cloud. Enterprises can unify AI training, RAG pipelines, high - throughput data processing, and unstructured data lakes on a single high - performance platform, enabling AI training, RAG, and inference across hybrid environments and bypassing the barriers of public - cloud providers.

It is worth noting that in addition to the deep partnerships with leading customers, this startup's OEM cooperation with Cisco, Supermicro, and HPE enables it to obtain lower hardware costs while maintaining software premiums, achieving the advantages of high gross margins and rapid customer acquisition.

With sufficient cash flow, foreign media also believes that VAST Data's new financing may indicate an acceleration of its pre - IPO preparation process. Last year, the startup also hired Amy Shapero, the former CFO of the global e - commerce platform Shopify.

04. Conclusion: Riding the AI wave, VAST Data's revenue soars

The importance of building AI infrastructure platforms such as data and computing power is increasing. Although technology giants such as NVIDIA, Microsoft, and Google dominate the top of this market with their GPUs and cloud platforms, the business growth and soaring valuations of AI startups represented by VAST Data prove that this is not just a game for giants.

In the AI competition surrounded