Key takeaways from the video:
- NVIDIA dominated AI GPUs (94% share in Q2 2025) thanks to its mature CUDA ecosystem, while AMD lagged at 6% with weaker software support.
- AMD invested heavily (Xilinx acquisition for $49B, Nod.AI) and released the MI300X, which outperformed NVIDIA’s H100 in memory bandwidth, latency, and cost-per-token at scale for large AI workloads.
- Despite early customer wins (Microsoft, Meta, Dell), MI300X revenue fell short of projections ($5B vs. expected $8B), hampered by ROCm’s usability gap versus CUDA and analyst skepticism.
- The game-changer: OpenAI’s massive partnership with AMD for next-gen Instinct GPUs (MI450 series onward, starting late 2026), potentially worth up to $100B; OpenAI helped design the architecture, and AMD issued warrants for up to 160M shares (10% of company) tied to milestones.
- This deal signals a potential break in NVIDIA’s monopoly, but AMD’s success hinges on closing the ROCm software gap before large-scale deployment.
Discover more from Erkan's Field Diary
Subscribe to get the latest posts sent to your email.