The AI GPU market has belonged to hyperscalers, but AMD’s MI350P is coming for the enterprise

Enterprise GPU access is expanding as Advanced Micro Devices Inc. lowers barriers with the air-cooled Instinct MI350P GPU and a ready-to-run software stack designed for standard enterprise servers.

The next wave of enterprise GPU adoption hinges not just on raw silicon performance, but on dramatically lowering the barrier to deployment. AMD is betting that pairing mainstream server form factors with polished out-of-the-box software will be the unlock that converts a massive untapped enterprise market, according to Mike Darby (pictured, right), senior manager of business development, Instinct Data Center GPUs, at AMD. That includes a low-commitment entry point through Dell Technologies Inc., which is making the hardware available for remote proof-of-concept testing before enterprises commit to a full deployment.

“The cool thing … is that [the XE9785 and XE9785L] are in Dell’s Customer Solution Center right now, live, ready to take remote POCs,” Darby said. “You don’t have to buy the huge server to try it and prove that it works. Dell has that ready to go live right now.”

Darby spoke with Gemma Allen (left) at Dell Technologies World 2026, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed enterprise GPU access, the new Instinct MI350P series and opportunities in on-premises AI infrastructure. (* Disclosure below.)

Enterprise GPU access expands with AMD Instinct MI350P

The centerpiece of AMD’s Dell Technologies World 2026 story is the new Instinct MI350P, a PCIe-based GPU designed to run in mainstream PowerEdge servers at 450 watts — well within the reach of standard air-cooled enterprise data centers. The MI350P offers 144 gigabytes of HBM3E memory, which is positioned as a key differentiator for inference performance given the critical role memory bandwidth plays in time-to-first-token latency, Darby explained.

“It’s a lot of memory, but it’s also HBM3E, so [it has] an amazing amount of bandwidth,” Darby said. “The main use case being inference — it’s all about that latency, time to first token, the user experience so that your attention span doesn’t move somewhere else while you’re waiting for your response.”

On the software side, AMD is pairing the MI350P with ROCm 7 and AMD Inference Microservices, giving enterprises a pre-optimized stack for the most popular open-weight AI models. The goal is eliminating the specialist expertise that has historically made GPU deployments inaccessible to mainstream IT organizations, Darby noted.

“What people are going to be excited about this year, especially in the enterprise space, is the out-of-box performance with ROCm 7, with AMD Inference Microservices,” he said. “You turn it on, launch the AIM, and you have out-of-box performance on the most popular AI models without having to have a PhD or computer science degree to figure it out.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of Dell Technologies World 2026:

(* Disclosure: AMD sponsored this segment of theCUBE. Neither AMD nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE