Advanced Micro Devices announced on July 28, 2026 that its MI350-class accelerators secured additional multi-year contracts with major cloud providers seeking to diversify AI GPU supply beyond NVIDIA. The wins do not dethrone NVIDIA, but they mark AMD's clearest commercial traction yet in large-scale training and inference clusters.
Why Hyperscalers Want a Second Source
Microsoft, Oracle, Meta, and other large buyers have spent two years absorbing NVIDIA allocation constraints. Dual-sourcing reduces concentration risk and improves negotiating leverage on price and delivery schedules. AMD's software stack has improved enough that some inference and fine-tuning workloads can migrate without full CUDA rewrites.
Limits of the Challenge
NVIDIA still dominates developer mindshare via CUDA and holds the majority of AI accelerator revenue. Analysts at Bernstein estimate AMD's data-centre GPU share remains well below NVIDIA's, even after MI350 wins. Custom ASICs from Google, Amazon, and Microsoft also compete for the same diversification budget.
Investor Takeaway
AMD shares often trade as a high-beta proxy for the AI semiconductor theme. Fresh cloud contracts support the bull case into second-half 2026, especially if HBM memory supply stays tight and NVIDIA cannot fill every rack. For NVIDIA's counter-narrative, see Blackwell Ultra order visibility. Live AMD quotes: Yahoo Finance.