Dell and NVIDIA Open Tech World 2026 with PowerEdge for 256 Blackwell Ultra GPUs per Rack
Michael Dell and Jensen Huang inaugurated Dell Technologies World in Las Vegas with the announcement of AI Factory 2.0: new PowerEdge servers support up to 192 Blackwell Ultra GPUs per chassis and 256 GPUs per rack, with LLM training up to four times faster than the previous generation.
Dell and NVIDIA Open Tech World 2026 with PowerEdge for 256 Blackwell Ultra GPUs per Rack
Michael Dell and Jensen Huang shared the stage at the opening of Dell Technologies World in Las Vegas on Monday morning to announce the second generation of the Dell AI Factory, a programme that combines servers, networking, storage, and NVIDIA's software stack into a reference platform for enterprises running AI workloads at scale.
The hardware highlight was the new PowerEdge XE9780 and XE9785 servers, with liquid-cooled versions (XE9780L and XE9785L). Each chassis supports up to 192 NVIDIA Blackwell Ultra GPUs, and Dell claims that the IR7000 rack can be configured with up to 256 GPUs. Compared to the previous generation PowerEdge XE9680, Dell projects language model training to be up to four times faster.
The NVIDIA AI Enterprise suite, which includes the vendor's frameworks and microservices, will now be sold directly by Dell, with its own managed services on top. This is a move towards the direction that SAP, ServiceNow, and Salesforce have been signalling for weeks: the delivery of AI as packaged infrastructure, rather than isolated proof-of-concept projects.
What Changes for Infrastructure Designers
The market reading is that Dell and NVIDIA are paving the transition from the prototyping phase of generative AI to the production phase, where latency, cost per token, and GPU density will dictate the architecture. For service providers and infrastructure consultancies, the opportunity window lies in deployment engineering: scaling liquid cooling, designing InfiniBand networks, integrating with PowerScale storage, and operating clusters in regulated environments.
The event runs until 21 May, with breakout sessions focused on data activation and operation of AI factories on 18 and 19 May.