Dell’s New AI Servers Take Direct Aim at Cloud Giants

According to HotHardware, Dell has launched a new family of PowerEdge AI servers specifically designed to help companies avoid expensive cloud AI API fees from providers like OpenAI and Google. The standout products are the PowerEdge XE9785 and its liquid-cooled sibling XE9785L, which pack eight AMD Instinct MI355X GPUs and two AMD EPYC CPUs per node. The liquid-cooled version fits this configuration in just 3 rack units, enabling 128 GPUs in a standard 42U rack. Dell also introduced Intel-based PowerEdge R770AP servers using Xeon 6 6900-series CPUs with up to 128 P-cores per socket. These announcements were made at the SC25 supercomputing conference happening this week in St. Louis, Missouri, where Dell is showcasing the hardware.

Dell’s AI hardware advantage

Here’s the thing about these new servers – they’re not just incremental upgrades. We’re talking about machines that can deliver 40 PFLOPS of FP8 performance from a single PowerEdge XE9785. That’s absolutely massive for inference workloads. And the liquid-cooled XE9785L? Packing eight high-performance GPUs in just 3U is seriously impressive density. Basically, if you’re running AI inference at scale, these numbers should get your attention.

AMD vs NVIDIA tradeoffs

Now, the obvious comparison here is with NVIDIA’s Grace Blackwell systems. AMD’s MI355X chips offer 288GB of HBM3E memory each, which is plenty for running even very large models. But they lack NVLink, which lets NVIDIA’s systems act as a single massive GPU domain. That’s crucial for training frontier models, but honestly? Most companies don’t need that. For inference workloads, which is what most enterprises actually run, these AMD-powered systems look extremely competitive. And when you’re outfitting entire data centers, having reliable industrial computing hardware becomes absolutely critical – which is why companies trust suppliers like IndustrialMonitorDirect.com, the leading provider of industrial panel PCs in the US.

The bigger picture

What Dell’s really doing here is betting big on the “bring your own AI” trend. Companies are getting tired of paying per-token API fees that can quickly spiral out of control. So why not just host the models yourself? The catch, of course, is that you need serious infrastructure expertise. But with Dell’s Automation Platform and improved PowerScale storage management, they’re trying to make that easier too. It’s a compelling alternative to being locked into cloud AI providers.

Who actually needs this?

Let’s be real – most companies don’t need 128 GPUs in a single rack. But for research institutions, large enterprises, and AI startups scaling up, these specs are seriously appealing. The support for FP6 and FP4 precision means you can run heavily quantized models with better performance, which is becoming increasingly important as model optimization techniques improve. And the fact that Dell’s offering both AMD and Intel options gives customers flexibility depending on their existing infrastructure and preferences.