AMD
Thesis
- The business: a fabless designer of x86, GPU and FPGA chips.
- Growth engine: AI datacenter spending (Instinct GPUs and Helios racks), server CPU share (EPYC).
- Moat: server CPU franchise, chiplet design and lead-customer standing at TSMC, co-design with frontier labs, Xilinx FPGA design-ins with long product lives, net-cash balance sheet.
- Risks: volume paid for in equity, credit exposure to customers, custom chip substitution, the NVIDIA Intel alliance, first generation rack execution (Helios), memory squeeze.
- What to monitor: first-gigawatt deliveries and warrant vesting, Mercury server share, and purchase commitments against Data Center revenue.
AMD versus NVIDIA
- The verdict: Nvidia has stronger growth, superior margin conversion, and stronger moat.
- NVIDIA's moat in CUDA (less strong with hyperscalers) and scale (R&D dollar).
- NVIDIA has stronger growth (hyperscale AI systems, AI clouds, sovereign and enterprise, networking attach). AMD could beat if MI450 land on time.
- NVIDIA has superior margin conversion on pricing power.
- Downside risk:
- In a digestion, AMD's AI ramp stalls first. NVIDIA's reported numbers fall harder but stay highly profitable. AMD's exposure is its competitor. When NVIDIA's supply loosens, buyers cut the second source first.
- Nvidia is more exposed to the financing loop.
- AMD is more exposed to 1) first-party rack execution, 2) ARM designs taking share in server CPU, 3) consumer market slowdown.
AMD versus Intel
- AMD grows on share gains and contracted AI demand while running a fabless cost structure; Intel's design franchise is now highly profitable again, but it carries a foundry that lost $10.3B in 2025 and must still win outside customers to earn its capital back.
- Risk: Intel has higher operating leverage. AMD is subject to AI demand concentrated in a few labs, and leading-edge wafers from TSMC.
- Growth: AMD's growth comes from share and new categories under contract; Intel's comes from price in a shortage. The first compounds, the second mean-reverts.
- Margin: AMD converts each dollar of sales into about 13 more cents of gross profit; closing that gap requires Intel Foundry customers, not Intel cost cuts.
- Risk: In a general downturn Intel breaks first on high fixed cost. AMD breaks first in an AI-capex pause or a Taiwan disruption.
- Moat: Intel's hardest-to-copy asset is its fabs, which currently destroy value; AMD's advantage is execution and riding on TSMC's best nodes, which is copyable in principle but has compounded for eight years. The moat race turns on whether Intel's manufacturing becomes an advantage (18A/14A at cost parity) before AMD's execution lead widens further.
- Supplier: Intel owns more of its supply chain; AMD has passed costs through better. Implication: Intel wins continuity in a Taiwan disruption; AMD wins on cost flexibility in every other scenario.
- To watch:
- Intel lands a definitive 14A anchor customer and Intel Foundry's loss shrinks toward breakeven by 2027, turning the fab base from a cost into a second profit engine with scarce US capacity.
- AMD's MI450 and Helios ramp slips or yields stay low, so 2027 AI growth falls short and the warrant-linked anchor customers delay purchases.
- Intel's 18A Xeons restore performance leadership and AMD's server share stalls below its >50% revenue-share target.
