Enterprise AI workloads are moving back on-prem because cost, data gravity, and revenue capture all push them in that direction. At Dell Tech World 2026, the company responded to this broad trend with an expanded Dell AI Factory and a deeper Dell AI Data Platform layered on top of the largest single-event set of announcements for storage and cyber resilience that Dell has ever staged. The combination reshapes the cloud-versus-on-prem decision for the next phase of enterprise AI.

The Forces Pulling AI Back On-Prem

Running production AI in the public cloud is getting too expensive, and the financial models that justified it are starting to break. Agents that run all day, token bills that scale with usage, and the cost of moving data in and out of inference now land directly on the CFO. According to the November 2025 Dell Technologies IT Strategy Pulse, 81% of IT and business decision makers say their organization is moving toward a disaggregated infrastructure model. That matters because the shift is toward hybrid deployments, with companies looking for resources they can scale independently and AI workloads they keep under their own control, inside environments they trust. AI inference in particular fits the on-prem side of that hybrid model.

Three forces are pushing production AI back on-prem. First, agents and inference workloads run continuously, and the bills come in bigger than anyone planned for. Second, data is hard to move, and most enterprise data still lives on-prem. Third, companies that buy AI services want costs they can predict and control, instead of metered cloud bills tied to usage they can’t fully plan around. There’s a fourth factor, which arises because storage sits underneath all the other considerations, though this factor tends to get buried when so much of the conversation focuses on GPUs. The same Dell research reports that 69% of decision makers see their existing storage capacity and capabilities as insufficient for meeting the growing demands of AI, which means storage is the next thing that slows AI down once companies have the GPUs in place.

Any on-prem stack that wins this next wave of enterprise AI has to deliver on four fronts. First, it needs a real data layer that lives above storage, with prep, curation, and vector indexing handled as primary workloads. Second, it needs cyber resilience built into the same package as the AI workload, bought and run through the same control plane. Third, it needs to host the same frontier models customers can run in the cloud, with the models running where the data lives. Fourth, it needs an orchestration layer that lets enterprise IT teams operate the whole stack without becoming hyperscalers themselves. This applies especially in regulated industries: For a bank that cannot let customer data leave the building, a hospital working under HIPAA, or a government agency, none of those four is optional. Miss one of the four and customers will go looking for a vendor that has them all.

Dell’s Move to a Complete On-Prem AI Stack

On the AI Factory side, Dell deepened its NVIDIA partnership with the launch of PowerRack, which turns the AI Factory into rack-scale infrastructure that customers can simply order across compute, networking, and storage. Dell Deskside Agentic AI puts agent runtimes on Grace Blackwell workstations, with the NVIDIA NemoClaw umbrella as the software stack and CrowdStrike as the security layer. The Dell AI Data Platform was reorganized into a four-layer stack mapped to the people who actually use it, with a unified user experience across the layers. The Dell AI Data Platform also became the storage foundation for NVIDIA Omniverse, and Dell now calls out physical AI as a third workload type alongside generative and agentic AI. The expanded Dell AI Ecosystem Program brought frontier models on-prem, including Gemini on Google Distributed Cloud, Grok via SpaceX, Reflection’s open frontier models, Hugging Face via the Dell Enterprise Hub, and Mistral, Cohere, Meta, and Poolside as core model partners. Palantir’s Foundry and AIP, along with ServiceNow Otto, landed as the application layer on top.

The modern datacenter track carried the biggest set of storage announcements Dell has stacked together in years, maybe ever. PowerStore Elite extended the midrange storage line. PowerProtect One unified Data Manager and Data Domain under one brand, and Cyber Detect added an AI engine that watches for data corruption and tells customers which backups are actually clean to restore from. Dell Exascale Storage put Lightning at the high-performance AI tier, where the parallel file system keeps data-prep pipelines and GPU clusters fed at extreme throughput. Dell Private Cloud broadened multi-hypervisor coverage across VMware Cloud Foundation 9.1, Nutanix on PowerStore, Microsoft Azure Local, and Red Hat, and the rebranded Dell Distributed Private Cloud (formerly NativeEdge) folded the edge into the same umbrella. Sitting above all of it is the Dell Automation Platform, the agentic ITOps control plane, now joined by a new Dell Automation Studio that lets customers build their own automation workflows.

The Sovereign Stack Dell Hasn’t Named

What Dell got right at DTW 2026 is the portfolio-level integration. The enterprise AI conversation has been stuck on the same problem since ChatGPT set off the AI boom in late 2022, namely that delivering a production AI workload requires stitching together too many vendors. Dell put every layer in one place. Bringing frontier models on-prem under customer control collapses the central trade-off that has hampered enterprise AI strategy. Enterprises have faced a binary choice between cloud-grade model capability and on-prem data control. The strongest models lived behind cloud APIs, while anything a company could run on its own infrastructure meant settling for weaker open models. Dell now puts Gemini, Grok, Reflection, Hugging Face, and the rest on customer infrastructure where the data already lives.

The AI Data Platform reorganization is the second structural win. Data prep, curation, and vectorization are the workloads actually driving AI capacity demand, and treating those workloads as a layer above storage moves Dell out of the trap of selling AI as a storage feature. It also aligns the pitch to the data engineer and the AI engineer, the actual buyers. Bundling cyber resilience with the AI workload addresses the new attack surface that AI creates, something the bolt-on protection model has never fully covered. Dell Private Cloud answers the VMware and Broadcom disruption with a hypervisor-neutral landing zone that pulls storage and AI infrastructure with it, and the Automation Platform on top makes the stack operable by an enterprise IT team rather than a team of platform engineers. The combined effect is that Dell now sells the integration itself. Customers buy a stack that already works together, with every product supporting that promise.

Inside that completeness sits a sovereign stack Dell hasn’t openly named or claimed yet. “Sovereign” here means that the customer controls the data, the models, the runtime, and the protection inside its own walls and under its own governance. Today this portfolio includes frontier models, an enterprise data layer, agent runtimes, cyber resilience, a multi-hypervisor private cloud, an automation layer for deployment and ongoing management, and the underlying hardware stack of compute, storage, and networking. Tied together, they deliver the secure, trusted, sovereign AI environment that regulated enterprises have been asking for.

Dell’s own slides from its DTW 2026 presentation list the reasons customers want AI on-prem (regulation, bandwidth, real-time processing, mission-critical work), and that list lines up almost exactly with how a sovereign stack gets sold. Other established vendors can also tell a credible on-prem AI story today, but Dell’s DTW announcement separates the company on the breadth of frontier models brought on-prem, with Gemini, Grok, Reflection’s frontier models, the Dell Enterprise Hub for Hugging Face, and a deep bench of open-weight partners now sitting on top of the rest of the portfolio. Each of those reaches the customer through a partnership rather than a Dell-built model, which is what makes the breadth possible. Dell is the integration platform that puts every one of these frontier models inside the customer’s own datacenter. That combination gives Dell the strongest claim to a complete on-prem AI stack in the market today.

So the capability is built. Now it’s time for the positioning to catch up. In my view, Dell has built more than its original sales motion knows how to sell. The data platform reaches a different buyer than the infrastructure audience the field has spent decades on, and Dell has openly recognized that gap. Closing that gap is the real test of the broader go-to-market push underway, and Dell is working it through several channels. Forward-deployed engineers who understand the data conversation, deeper partnerships with system integrators like WWT and Presidio, and tighter integration with the data and analytics ISVs all sit in the program.

That mix is the right shape to fit the problem, but the success of the execution will only become visible over multiple quarters. Among other hurdles to clear: Knowledge graphs inside the AI Data Platform are still on the roadmap, and the structured-context advantage only counts if Dell builds the knowledge graph natively inside the platform, with no partner standing in for it. Also, the PowerStore midrange recovery still depends on the Elite generation of products winning share back from incumbent midrange rivals.

Where Cost, Data, and Revenue Take Enterprise AI Next

Where to run an enterprise AI workload is no longer just a question of aiming it at the cheapest GPUs. Rather, it sits where cost, data location, and revenue capture meet, and every one of those forces pulls the workload toward infrastructure the enterprise owns and runs itself. Companies running production AI want the data prep, the model, the protection, and the orchestration to feel like one environment, instead of four separate purchases stapled together. The on-prem stack that wins the next phase will be the one that delivers that integration without forcing customers to give up choice on hypervisor, on data ISVs, or on model providers. Enterprises want flexibility and control in the same purchase, and they will pay for the vendor that delivers both.

Dell now delivers both. Every layer of the AI workload sits in the portfolio, with the integration done and the customer’s choices left intact, and the result is the most complete on-prem enterprise AI stack the company has ever shipped. The work that remains is naming the positioning out loud and letting the field motion catch up to the architecture. The capability is already in place. Everything else follows. For enterprise buyers rethinking their AI economics, the cloud-default assumption is worth another look now that this much of the on-prem stack sits inside one vendor.