The cloud-only AI narrative is getting harder to defend. At this year's Dell Technologies World in Las Vegas, Dell is making its case – backed by real customer deployments and new product announcements – that enterprises are running serious AI workloads on their own infrastructure, from corporate data centers down to individual workstations.

Dell also took the opportunity to roll out meaningful updates to the "AI Factory" platform it launched with Nvidia two years ago, all aimed at solving one of the industry's most persistent challenges: actually getting AI from proof-of-concept into production.

The customer traction is real. Dell now counts over 5,000 enterprises as AI Factory customers, all running Nvidia GPU-equipped servers alongside their cloud resources. The hybrid model these companies are adopting is a deliberate architecture: local compute reduces latency, keeps sensitive data off public networks, and increasingly, makes strong economic sense.

For certain workloads, Dell claims on-prem solutions can be up to 63% more cost-effective than relying on the public cloud. Those are Dell's figures, so apply appropriate skepticism, but the underlying trend is real and growing. Cloud AI compute capacity constraints are also becoming a legitimate concern for large enterprises, another reason organizations are finding value in owning a portion of their own token-generating infrastructure.

On the hardware side, Dell introduced a refreshed PowerRack lineup combining new PowerScale storage options with updated high-speed networking, a more complete, pre-integrated foundation for enterprise AI systems, similar in spirit to what Nvidia sho

wcased at GTC earlier this year. New cooling solutions and an updated

rack controller round out the picture.

More interesting, from a differentiation standpoint, is Dell's push into what it calls "deskside" AI: running local LLMs directly on workstations using Nvidia's NemoClaw software.

Local inference can improve experimentation speed, privacy, and autonomy. While it may initially sound like little more than a niche workstation story, it's actually a strategic part of hybrid AI. For heavy AI users like developers, researchers, and agent builders, the economics are striking: Dell claims payback periods as short as three months and cost savings versus public cloud APIs of up to 87%.

Again, these are Dell's own numbers, but even discounted they reflect a meaningful shift in how organizations are thinking about where AI compute should live. More importantly, deskside AI completes what Dell is positioning as a true three-tier hybrid architecture: on-device, on-prem, and cloud – with each layer handling the workloads it's best suited for.

On the software side, the most significant update was to Dell's AI Data Orchestration engine, which uses tools like Nvidia NIMs to help organizations turn their proprietary data into productive AI workflows.
<

br/> Dell also deepened its ties to Nvidia's Omniverse platfo

rm for physical AI and digital twin applications, and expanded support for Nvidia's cuDF analytics engine. These updates collectively flesh out what Dell calls its AI Data Platform, the software layer designed to tie the whole hybrid architecture together.

Services may be the least glamorous part of Dell's story, but they're arguably the most important. One of the clearer lessons from the past two years of enterprise AI deployments is that most organizations need significant help translating infrastructure investments into working solutions.

Dell has leaned into this, developing blueprints and best practices, many of them tailored to specific industries, that can meaningfully improve the odds of AI projects actually reaching production.

In the case of Dell's deskside AI offerings, it's the services element that really differentiates them from what other hardware vendors are bringing to market.

A range of new certified solutions built with partners including ServiceNow, Mistral, CrowdStrike, and Uneeq, many of them integrated through Dell's Automation Platform, extend this approach further.

The model av

ailability question is also worth addressing directly. The best frontier mode

ls will remain cloud-based for the foreseeable future, but the gap is closing faster than most people expected.

Dell announced new partnerships to run Google's Gemini 3 Flash on Dell infrastructure via Google Distributed Cloud, along with SpaceX's Grok models, and open-source options from Hugging Face. Equally important: not every AI workload needs frontier-model capabilities. Many production use cases run perfectly well on smaller models, the kind that fit comfortably on-prem or even on a workstation.

The intelligent orchestration of workloads across all three tiers, deciding in real time what runs where, remains an unsolved problem and an important one. But the trajectory is clear.

Compute is getting faster, models are getting smaller and more capable, and the services infrastructure needed to make hybrid AI actually work is maturing quickly. Dell still has work to do to help more customers move from pilot to production, but the momentum is genuine, and the case for hybrid AI as the dominant enterprise architecture is getting stronger by the quarter.