**The most surprising finding is that the big AI-surge names are not the big winners. The biggest price multiples come from small Korean and Japanese equipment makers.** The per-stock table (step 1) shows the minimum and maximum close for the ten Korean and Japanese stocks in the dataset: - **Lasertec (6920.T):** the close runs from about 100.6 to 44,520, a max/min ratio of about **443×**. - **HANMI Semiconductor (042700.KS):** about 661 to 187,505, roughly **284×**. - **Advantest:** about **85×**. - **Disco:** about **75×**. - **Tokyo Electron:** about **50×**. - **SK Hynix:** about **17×**. - **Samsung:** only about **5.6×**. Samsung is the largest and most recognisable semiconductor stock in this set, yet it is the weakest of the ten on this measure. Lasertec's multiple is roughly 79 times Samsung's. The stocks that look like the obvious AI winners barely register next to the niche tool and equipment makers. Hanmi, for example, makes chip-packaging equipment. A few cautions apply: - **It is not a return:** the ratio compares the lowest and highest close over 2012–2025. It ignores the order of the two points and any dividends or splits. - **Prices are in local currency:** SK Hynix's maximum close is exactly 300,000.0, a suspiciously round number that I did not verify. Raw price levels are not comparable across exchanges, so only ratios like these mean anything. - **Some price rows look unreliable (step 0):** - 461 of 422,729 rows have a close outside the day's high–low range. - 10 rows have an open outside the range. - 4,888 rows have zero volume. - 6,840 rows are flat days. - The minimum close anywhere in the data is 0.0093. - **No duplicates:** there are no duplicate (stock, date) keys. - **Stock count:** the data holds 152 stocks, slightly more than the roughly 149 the card lists. Because of these issues, I'd treat the extreme multiples, especially Lasertec's and Hanmi's, as a pattern to verify against the price history rather than a settled fact.