When we first extended Qumulo’s data platform into the public cloud, our thesis was straightforward. We believed enterprises would use the cloud as a natural extension of their data centers, primarily for two low-intensity workloads: backup of unstructured data in a hybrid model, and long-term archiving using services like Glacier Instant Retrieval, paired with Cloud Native Qumulo’s cold tier, to provide a highly durable but cost-effective way to keep data always online. The assumption was that the cloud would be a cheap, resilient safety net for the less critical classes of enterprise data.
The reality that emerged has been the inverse of that expectation. Today, over half of our cloud customers are not starting with backup or archival workloads at all. They are beginning with the hot tier—their most valuable, performance sensitive, business critical data. These are the datasets enterprises depend on for innovation, analytics, and AI reasoning. The reason is simple: in the cloud, data is not just inert storage, it is the fuel for computation. The public cloud offers a utility model of instantly available, globally distributed, and infinitely elastic computational capability. Instead of waiting months to stand up specialized infrastructure—data centers, networks, and exotic compute configurations—customers can bring their most valuable data into the cloud and place it directly beside GPU clusters, AI accelerators, and hyperscale analytic engines that are available on demand.
This behavioral inversion has reshaped how enterprises think about hybrid and multi-cloud strategy. We are now watching customers use our Cloud Data Fabric to actively stream data from their on-premises data centers into three or four distinct public clouds simultaneously. They aren’t doing this for redundancy; they are doing it to selectively exploit the differentiated computational strengths of each cloud provider. A financial services firm may choose Azure for regulatory-grade analytics, AWS for AI training, GCP for data science workflows, and OCI for HPC—all at once—because Qumulo can put their data within reach of each of those environments without the enterprise losing sovereignty or control.
This perspective stands in sharp contrast to voices like Michael Dell, who in 2018 asserted that “as much as 80 percent of the customers across all segments … are reporting that they’re repatriating workloads back to on-premise systems because of cost, performance and security.” While that claim may have held rhetorical power in the early wave of cloud adoption, it mischaracterizes the nature of what is actually being repatriated. In our conversations with customers and industry peers, the workloads that tend to return to on-premises environments are not the hot, computationally intensive datasets. They are backups. By design, backups are cold, static, and computationally inert—storage-intensive but not performance-intensive. Pulling those workloads back is, at best, a cost optimization exercise, not evidence of a mass exodus of enterprise innovation from the public cloud.
Even more recently, Dell has argued that “AI inferencing on premises, where 83% of the data already is, is 75% more cost-effective than relying on someone else’s public cloud.” The problem with this statement is that it conflates the location of stored data with where value is actually created. Enterprises are not moving their most valuable datasets into the cloud to lower their storage costs; they are doing it to unlock immediate access to elastic compute at a global scale. The cloud has become the place where the most innovative workloads live precisely because computation—not storage—is the scarce and differentiating resource. Cost effectiveness cannot be reduced to pennies per gigabyte; it must be measured by the speed at which enterprises can reason across their data and act.
The inversion of our original thesis underscores a profound market truth: the cloud is not merely a storage endpoint. It is a computational frontier. Data that matters most will migrate to where it can be acted upon most powerfully. That is why Qumulo’s architecture—any data, any location, total control—is resonating so strongly across industries. Our customers are not looking at cloud versus on-premises as a binary choice. They are embracing a fluid data fabric where value intensive workloads flow to the clouds best suited to them, while cost sensitive archives find a durable home in hybrid or on-premises environments.
This is the new shape of enterprise IT. The future belongs not to static data silos, but to universal data fabrics where active datasets move at the speed of innovation, computation is consumed as a utility, and the most valuable data resides in the places where enterprises can extract the most value. That future is here today, and Qumulo is proud to be the platform that makes it possible.
