AMD Aims for “Milan Moment” With Instinct MI450, Strengthening Software Stack to Compete With NVIDIA

AMD is preparing for what it sees as a pivotal moment in its AI accelerator strategy. At the Goldman Sachs Communacopia Technology Conference, Forrest Norrod, AMD’s EVP of Data Center Solutions, emphasized the company’s commitment to ensuring that the upcoming Instinct MI450 lineup won’t just compete with NVIDIA on hardware - but also on the critical software stack that drives AI adoption. Transcript link.

Norrod was clear:

“We’ve focused on getting there in the 450s so that for training, there’s no excuses or impediments. There’s no hesitation of, ‘hey, if I’m training, I’ll be behind this generation if I go with AMD.’ And that’s been the realization.”

This signals that AMD has learned from its MI300 series, which delivered strong inference performance but lagged behind NVIDIA in training adoption due to slower time-to-introduction for certain software features.

Norrod went further, describing the MI450 as AMD’s “Milan moment.” This is a reference to the EPYC Milan CPU generation, which marked a turning point for AMD’s server business, allowing it to claw significant market share away from Intel. By making this comparison, AMD is positioning the MI450 as the product that could redefine its role in the AI accelerator space.

The MI450 lineup is expected to go head-to-head with NVIDIA’s next-gen Vera Rubin architecture, and AMD is eager to ensure customers won’t hesitate when weighing their options. Central to this strategy is a stronger software ecosystem, with improvements in ROCm, expanded framework support, and developer tooling meant to eliminate adoption friction.

Technical expectations for the MI400 lineup include:

• HBM4 integration with capacities up to 432 GB, delivering massive bandwidth improvements

• Aggressive rack-scale expansion, including the upcoming Helios rack, designed to rival NVIDIA’s top Vera Rubin configurations

• Targeting not just inference, but robust training workloads - AMD’s key growth area for AI adoption

With MI450, AMD is signaling that it doesn’t want to be a secondary option in the AI accelerator race. Instead, it aims to deliver a hardware-software package strong enough to go toe-to-toe with NVIDIA, potentially shifting enterprise and hyperscaler sentiment in AMD’s favor - just as EPYC Milan did for CPUs.

Do you think AMD can replicate its EPYC Milan success in the AI accelerator space with the Instinct MI450, or is NVIDIA’s ecosystem too entrenched?