During her keynote at Computex 2025, AMD CEO Lisa Su announced a strategic expansion into high-performance AI infrastructure with the launch of the upcoming MI450 GPU and the associated AMD Instinct server platform.

These new chips will serve as AMD’s answer to Nvidia’s H100 and B100 accelerators, which currently dominate the training and inference of large AI models. AMD claims the MI450 will deliver 2.5× performance over its current MI300X chip and will begin shipping to partners in mid-2026.

What makes this announcement even more impactful is the revelation that OpenAI is already testing and validating AMD’s new chips, a clear indication that the AI ecosystem is beginning to diversify its hardware stack—moving away from total Nvidia dependence.

Key Highlights:

• MI450 GPU Launching in 2026: Designed for high-performance AI training and inference workloads.

• 2.5× Performance Leap: Compared to AMD’s previous-generation MI300X GPU.

• OpenAI Testing: AMD confirmed OpenAI is among its early partners for AI model training.

• Direct Nvidia Competition: Signals a deeper challenge to Nvidia’s dominance in AI data centers.

• AI Server Strategy: Part of AMD’s broader push to become a key player in enterprise AI infrastructure.

Source:
Reuters / AMD Computex Keynote Coverage