**The most counterintuitive finding: in a dataset about the "AI surge", some of the biggest price swings were collapses, not gains. Wolfspeed and Kalray each ended about 99% below their peak.** **Why it's surprising** - The obvious way to read this dataset is to rank stocks by max-to-min close ratio. On that measure Wolfspeed looks like a huge winner, with a 354.7x ratio behind only NVIDIA (687.2x) and Lasertec (442.6x). - The dates show the opposite. Wolfspeed peaked on 2021-11-16 at about $141.87 and hit its low of $0.40 on 2025-06-30. Its last close is only 1.1% of its peak (step 8). - Kalray is the same story, with a 116.9x ratio. It peaked on 2021-02-08 at about €49.10, bottomed on 2025-05-20 at about €0.42, and closed at 1.3% of its peak (step 8). - BrainChip (ratio 229.9x) also peaked in January 2022 and sits at 9.4% of that peak. - The winners look very different. NVIDIA's low was in 2012 and its high on 2025-07-30, so it finished at 96.9% of its peak. Broadcom's low was in 2012, its high was on 2025-07-30, and it finished at 95.4% of its peak. - Lasertec (32% of peak) and Aehr (31% of peak) peaked in 2024 and 2023 and have since given back two-thirds or more of their gains. **Takeaway** Max/min ratios hide direction. In this dataset, the top of that ranking mixes the biggest winners with the biggest busts. You need the dates of the extremes and the last close versus the peak to tell them apart, so the "surge" was highly concentrated rather than sector-wide. **Data caveats** - Prices are in local currencies (KRW, JPY and others), so raw price levels are not comparable across companies. SK Hynix's maximum close is exactly 300,000.0, a suspiciously round number (step 0). - Step 0 also flags 461 rows where the close falls outside the high–low range, 4,888 zero-volume rows, and 6,840 flat days. - There are no duplicate stock/date keys, and no non-positive prices.