BN · rank #2 · 2026-10-21
AMD
NASDAQ · $872.95B (USD)
The exact numbers the algorithm saw.
| Composite score Z-score blend of the factors below; drives the rank. | 0.93 |
|---|---|
| Forward revenue growth Consensus forward revenue growth. | +74.9% |
| Forward net margin Consensus forward net margin. | +18.5% |
| Net margin TTM Trailing twelve month net margin. | +15.6% |
| Margin expansion Forward minus trailing net margin (percentage points). | +2.9% |
| Forward PEG Forward P/E to growth. Below 1 is cheap for the growth. | 0.63 |
| Debt / FCF Net debt relative to free cash flow. Lower is safer. | 0.51× |
| Analyst upside Spread between the consensus 12m target and the current price. | +18.2% |
| Last EPS surprise Most recent reported EPS versus consensus. | +23% |
| Market cap (USD) | $872.95B |
The AI research card
Independent qualitative review of each pick before the order is placed. On the rare day the research service is unavailable, the paper book trades on the quant ranking alone and no card appears here.
Summary
AMD designs and sells CPUs, GPUs, and AI accelerators spanning data center, client, and embedded markets, with its Instinct/EPYC Data Center franchise now the dominant revenue driver growing 107% YoY to $6.7B in Q2 2026.
Rationale
AMD's Helios rack (72x MI455X, 31TB HBM4, 2.9 exaFLOPS FP4) directly embodies the bottleneck-solver thesis—integrating scarce compute, memory bandwidth, and networking into a deployable system—while a 0.63 forward PEG against 75% revenue growth suggests the market is underpricing the structural demand capture.
Material risks
- 1Hyperscalers (Google TPU, AWS Trainium, Microsoft Maia) accelerating custom ASIC deployment could erode AMD's accelerator TAM from its largest customers, directly disconfirming the bottleneck-solver moat by internalizing the capability AMD is supposed to uniquely provide.
- 2NVIDIA's entrenched CUDA ecosystem and integrated NVL rack systems create a software-lock switching cost that AMD's open ROCm stack has not yet demonstrably overcome at scale, meaning benchmark wins may not convert to durable revenue share.
AI verdict council
Each pick is reviewed independently by 3 models before any order. 2 of 3 voted to proceed.