Tech teams have spent the past three years obsessing over AI models, GPUs, and the race to build ever-larger systems. But beneath the headlines, another battle is taking place that may prove equally as important.
Organizations are generating more data than ever. But the challenge is storing, accessing, and moving it efficiently enough to keep applications, analytics platforms, and AI systems operating at scale.
At the recent IT Press Tour in Boston, I learned how an open-source distributed filesystem company called JuiceFS is betting heavily on the simple idea that object storage has already won the storage war and is working on what comes next.
Object Storage Is Everywhere
Many people outside infrastructure teams rarely think about object storage, yet it has become one of the foundations of computing.
When companies store images, videos, training datasets, backups, documents, or application data in cloud environments, there is a good chance that information ultimately resides in an object store such as Amazon S3, Azure Blob Storage, Google Cloud Storage, or one of the growing number of S3-compatible alternatives.
Object storage offers exceptional durability, virtually unlimited scalability, and geographic distribution, all for relatively low costs. For many organizations, it has become the default destination for unstructured data. But a growing number of modern platforms are now built directly on top of object storage. Vector databases, AI platforms, analytics engines, search systems, and streaming architectures increasingly use object stores as their foundational data layer.
This trend has accelerated to the point where Amazon itself recently introduced Amazon S3 Files, allowing customers to mount object storage as a traditional file system. The launch serves as an acknowledgment that customers want the economics of object storage while retaining the simplicity of familiar file-based workflows. The problem is that object storage was never designed to behave like a file system.
Object storage excels at storing data, but applications often expect something different. Traditional applications were built around file systems that support directories, shared access, metadata operations, file locking, and POSIX compatibility. Developers expect to be able to rename files, update content, browse folder structures, and perform countless small operations without worrying about what is happening beneath the surface.
One example discussed during the presentation involved a large video file. Adding a small signature to the end of a traditional file is straightforward. In many object storage architectures, modifying a file may require reading the entire object, making the change, and writing it back.
As datasets grow from gigabytes into petabytes, those inefficiencies begin to matter. This is the gap JuiceFS is attempting to close.
A Different Architectural Approach
JuiceFS was founded in 2017 by a team with deep experience in distributed systems and data platforms. The company originally set out to address challenges associated with Hadoop Distributed File System environments, but its focus has broadened significantly as cloud-native infrastructure has matured.
Rather than treating every file as a single object, JuiceFS breaks files into smaller blocks and stores metadata separately from the underlying data. This allows the platform to present applications with a familiar file system interface while continuing to use low-cost object storage. The distinction may sound subtle, but it has important consequences.
Organizations gain the economics of object storage while preserving much of the behavior applications expect from traditional file systems. The result is a platform that attempts to bridge two worlds that have often been difficult to reconcile.
Why AI Makes This More Relevant
Like many infrastructure companies today, JuiceFS benefits from AI without positioning itself as an AI company. The reason is that AI workloads create enormous volumes of data. Training datasets contain billions of files. Inference environments need rapid access to shared information. Research organizations move petabytes of data between locations. Large language models depend on storage systems that can keep pace with increasingly demanding workloads.
While industry attention often focuses on compute power, many organizations are discovering that storage and data access can become bottlenecks long before they run out of GPUs.
One of the more compelling examples discussed during the session involved JuiceFS's mirror filesystem capability. The technology enables data to remain accessible across multiple regions while reducing the delays that often occur when compute and storage are separated.
For organizations operating AI infrastructure across multiple clouds or geographic regions, that capability becomes increasingly important.
Real-World Scale
What makes JuiceFS particularly interesting is that its story extends well beyond theory. The company reports that its community edition now manages more than an exabyte of data across hundreds of thousands of deployments and hundreds of billions of files. Its customer base spans AI companies, research organizations, cloud-native businesses, and enterprises managing large-scale analytics workloads.
Examples discussed during the presentation included South Korean technology giant Naver, AI company MiniMax, travel platform Trip.com, smartphone manufacturer Xiaomi, and Fly.io. These deployments show how JuiceFS is not solving a niche storage problem but addresses a challenge that arises when organizations attempt to combine cloud economics with file-based applications.
Perhaps the most interesting aspect of the JuiceFS story is its willingness to compete in a crowded and highly competitive market. The company openly identifies Weka, Luster, Panasas, and Amazon's new S3 Files offering as competitors.
Although this might look like an ambitious list, JuiceFS approaches the market from a different angle. Rather than selling premium proprietary infrastructure, it is attempting to give organizations greater flexibility by allowing them to continue using their preferred object storage platforms.
The Bigger Picture
The most important lesson from JuiceFS may have little to do with filesystems at all. For years, storage was often treated as background infrastructure. As long as data remained available, few people paid much attention to how it was stored. That assumption is becoming harder to maintain.
As AI systems, analytics platforms, and cloud-native applications continue to scale, the architecture sitting beneath those workloads is becoming increasingly important. There is no denying that storage decisions affect performance, costs, resilience, flexibility, and even the viability of future AI initiatives.
JuiceFS has built its business around a belief that object storage will continue to dominate modern infrastructure. If that prediction proves correct, the next phase of competition may not be about object storage itself. Instead, it will center on which platforms can make object storage accessible, efficient, and practical for the applications that depend on it.
JuiceFS is fighting this battle today, and it's also one that many technology leaders may soon find themselves paying much closer attention to.
Thank you to Alex Zaharov-Reutt for sharing the iTWire TV Recording of the JuiceFS presentation at the IT Press Tour in Boston.
I am hoping to get the team from JuiceFSon the podcast in the next few weeks. If there are any questions you would like me to ask, please leave a voicemail below to be a part of the conversation.