AMD (AMD) NASDAQ Investor Conference summary

AI product and technology roadmap

• Announced leadership CPU, GPU, and NPU architectures, offering end-to-end solutions for data center, PC, and embedded markets.

• Ryzen AI 300 Series for premium notebooks features 50 TOPS, surpassing competitors and enabling Copilot Plus AI PCs, available July.

• Previewed Gen 5 EPYC CPU servers (Turin) with improved performance per watt and dollar, launching in H2 2024.

• Annual GPU AI accelerator roadmap includes MI325 (2024), MI350 (2025, 35x performance improvement), and MI400 (2026), all with high memory capacity.

• Chiplet architecture and open interconnect standards (UALink) support scalability and flexibility for future products.

Ecosystem, networking, and software

• UALink consortium (8 major tech companies) to launch open interconnect standard in Q3, enabling up to 1,000 GPUs per network.

• Pensando acquisition strengthens DPU and networking capabilities, with industry-wide push for Ethernet as the dominant data center networking technology.

• ROCm 6.1 software stack enables efficient model porting from CUDA, with most models running out-of-the-box and porting times reduced to days or weeks.

• Porting complexity is no longer a major barrier, and ongoing improvements in software and tools are expected.

Market positioning and business mix

• Revenue is currently more inference-weighted due to later market entry and high memory capacity of MI300X, which lowers TCO for inference.

• Customers prefer MI300X for inference applications due to faster processing and cost efficiency.

• Supply chain tightness for HBM and CoWoS persists in H1, but capacity is expected to improve in H2 as partnerships with all three memory suppliers expand.