Artificial intelligence (AI) is set to significantly increase the demands on cloud infrastructure, making it “heavier” than ever before.
For years, cloud services and private networks managed relatively modest amounts of data, mostly in gigabytes and terabytes. However, with the rise of AI and deep learning, enormous volumes of data—such as photos, video, sound, and natural language—are now being processed. As a result, data that once fit comfortably within the gigabyte and terabyte range is now being measured in petabytes and exabytes.
This shift requires information systems, including cloud platforms, to scale rapidly in order to store and manage such massive amounts of data. However, the real challenge lies not just in storage but also in accessing that data at much higher speeds and, most importantly, at a significantly lower operating cost.
Some companies are already working on building the next generation of infrastructure to meet these needs. CoreWeave, a cloud computing provider that specializes in offering access to Nvidia’s advanced AI chips, is one of the companies at the forefront of this emerging market. In May, CoreWeave raised $1.1 billion in equity funding, bringing its valuation to $19 billion. Additionally, it secured $7.5 billion in debt financing from major investors such as Blackstone, Carlyle Group, BlackRock, and Nvidia itself.
CoreWeave, in turn, relies on a startup called VAST Data, which is revolutionizing the approach to cloud and private-network modernization with its software. VAST has developed what it calls a faster, more cost-effective, and scalable operating system designed to handle a variety of distributed networks.
According to Renen Hallak, CEO and founder of VAST Data, the company’s goal was to create infrastructure tailored to the needs of new AI workloads. Founded in Israel in 2016, VAST raised $118 million in a Series E funding round in December, nearly tripling its valuation to $9.1 billion. The company has also surpassed $200 million in annual recurring revenue, with a gross margin close to 90%.
Historically, data storage has been organized into tiers, where high-priority, recent data is easily accessible, while older data is placed further down, making it harder to retrieve. However, Hallak explains that this system doesn’t work for AI workloads.
“With AI, once you have a good model, you want to analyze all of your historical data because that’s how you generate value. And as you acquire more information, you need to retrain and improve the model,” he said. In this new era, data is repeatedly accessed across vast amounts of storage—sometimes reaching exabytes—making the problem vastly different from traditional data storage challenges.
Traditional systems often expand by adding nodes that store portions of the larger data set. However, this architecture requires constant communication between all parts of the system, and a single node failure can disrupt the entire structure. As a result, many enterprise systems are only able to scale to a few dozen nodes, which is insufficient for AI-driven workloads. In contrast, VAST’s approach gives every node access to all the information simultaneously, which enhances scalability, speed, and system resilience. Additionally, VAST unbundles the pricing for data storage and computing, which helps reduce costs.
While it might first seem that such advanced infrastructure is necessary only for tech giants, the demand for it is expanding throughout the economy. The shift is already underway in industries with highly data-intensive needs, such as animation. Pixar, the Disney-owned animation studio behind films like Inside Out 2, has been working with VAST Data since 2018.
Pixar’s adoption of volumetric animation, which produces more detailed characters and movements, has further intensified its data demands. Starting with the 2020 film Soul, Pixar has used this technique extensively, and it was even more pronounced in their 2023 release Elemental, where AI was utilized to curate the protagonist Ember Lumen’s flames.
The data demands for Inside Out 2 were significantly higher than those for Soul, requiring about 75% more computational power. Pixar’s previous method of moving data between high-performance and lower-performance drives when not in use no longer worked for rendering volumetric characters, says Eric Bermender, head of data infrastructure at Pixar. For AI-related tasks, Pixar now prefers on-premises networks over cloud-based solutions.
Bermender explained that AI workflows require processing vast amounts of diverse data that is neither cacheable nor sequential, which would traditionally be stored in lower-performance archive tiers. This shift challenges traditional architectures that are ill-equipped to handle the new data-intensive demands of AI.
For companies looking to adopt AI, the message is clear: they need to transition to a technology environment capable of managing AI’s unprecedented data requirements. This transition is comparable to the shift from gasoline-powered cars to electric vehicles, where every component must be reconsidered. Just as electric cars require new tires, AI needs an entirely new technological infrastructure to succeed.