Two Distinct Regimes
AMD's earnings reactions have meaningfully different statistical profiles before and after the AI-capex cycle inflected. Pre-2024, the median earnings-day |move| was approximately 6% — typical for a large-cap semi. Post-2024, the median |move| is approximately 9% — a 50% increase in event volatility, driven entirely by the AI overhang on each print.
The hold rates also diverge. Pre-2024, the 5-day hold rate was 5 of 6 (83%) but the median move was small enough that the hold/fade distinction didn't matter much. Post-2024, the hold rate dropped to 5 of 8 (63%) but the moves got large enough that the fade losses are material.
The AI-Cycle Reaction Pattern
Within the post-2024 cluster, two earnings reactions stand out: Q3 2024 (+11% on data-center GPU upside) and Q1 2025 (-9% on margin guide-down despite top-line beat). The cohort retrieval matches both kinds of setups depending on the post-call chart shape.
The single strongest predictor of follow-through across the 8 post-2024 prints is data-center GPU revenue YoY growth in the quarter. Above 90% YoY: 5 of 5 held the upside reaction. Below 70% YoY: 2 of 3 faded. The cohort retrieval doesn't see the YoY number directly — but it does see the post-call price reaction, which encodes the market's interpretation of the data-center number.
- Pre-2024 median earnings-day |move|: ~6%; Post-2024: ~9%
- Post-2024 hold rate: 5/8 (63%) at day +5; 4/8 (50%) at day +10
- Data-center GPU YoY > 90%: 5/5 held upside. < 70%: 2/3 faded
Reading Tuesday's Print
The most useful retrieval input post-AMD-earnings is the 9:30-10:30am ET chart shape on Wednesday — by that point the institutional flow has digested the call and the price action reflects the durable read, not the AH overshoot. Cohort retrievals against the 1h Wednesday chart match cleanly to the AI-cycle cluster; retrievals against the AH 5-min tape are noisy.