Advanced Micro Devices, Inc.
Information Technology · Semiconductors
AMD is the only credible second source to $NVDA in merchant AI accelerators and the share-taker against $INTC in x86 server CPU. Data Center segment now >50% of revenue; Instinct GPU run-rate scaled from ~$5B (2024) to multi-tens-of-billions on MI300X / MI325X / MI350 ramps with hyperscaler anchor commitments.
ROCm software gap vs CUDA remains the gating constraint on share take.
- EPYC server CPU share ~35-40% units, still climbing vs $INTC Granite Rapids
- Instinct MI350 (3nm, HBM3E) shipping H2 2026; MI400 series on roadmap for 2027 rack-scale
- $MSFT, $META, $ORCL, $OPEN-AI named as anchor MI300/MI325 customers - concentration but credible
- Xilinx (FPGA) + Pensando (DPU) give adaptive-compute optionality in networking + edge
- All wafers at $TSM N3/N5 - same node access as $NVDA, no fab capex drag
- ROCm vs CUDA software moat - accelerator TAM share still single-digit % of $NVDA
- Gross margin dilution as Instinct mix grows (lower margin than Ryzen/EPYC at maturity)
- Client + Gaming segments cyclical, embedded (Xilinx) in extended trough
- Customer concentration: top 3 hyperscalers >40% of DC revenue
- China export controls cap MI308 SKU revenue; further tightening = direct hit