The STAR 50 closed at 2,207.86 on June 30, 2026, after touching 2,255.25 intraday the next day. By October 30 it closed at 1,160.84, a 47.4% fall; its main ETF (588000) shows almost the same drop, 2.344 to 1.228. A Reuters analysis on October 26, when the index was already about 40% down, put the market value erased at roughly $480 billion and gave the piece the headline "In China, the AI bubble has already burst." Over the same window the SSE Composite fell about 12% (4,094 to 3,599). But this was not an isolated STAR event either: the CSI 300 lost about 21% from its June peak and Shenzhen's ChiNext about 45%. The damage tracked one thing, how much of each index was a bet on AI hardware. The STAR 50 sat at the far end: semiconductors were over 83% of its weight in July (Huxiu), and its June 12 reconstitution had just added Hua Hong, Yuanjie, Moore Threads and MetaX, after which the index rose another 33% in under three weeks to the top. The decline came in two legs. July took 26% off as global AI enthusiasm cracked. The doubts were about whether AI capital spending would earn its return, and in late July they were sharpened by reports that Nvidia might provide a roughly $250 billion backstop for an OpenAI data-centre project, the same worry behind a late-July rout in Korean chip stocks. August bounced 2.8%. Then September (-9%) and October (-24%) did the rest. Reuters found…
The revised text is tighter, and its best step is still the arithmetic. A multiple falling about 60% while the price falls about 40-47% is a clear way to show that the market stopped paying for earnings rather than earnings collapsing, and the post keeps the two readings side by side. The main weakness is that the earnings side rests on thin support. The post says the price and multiple moves "can only mean" trailing earnings rose about 40%. That is a derived figure from one Reuters article, corroborated by Chinese data points, and it is never checked against an actual trailing earnings…
The numbers are now the strongest part. The June 30 peak, the month-by-month legs, the Reuters P/E pair checked against Chinese data, and the arithmetic showing trailing earnings rose about 40% together settle that this was a de-rating rather than an earnings collapse. Using the yield's October retreat to rule out rates as the main driver is the kind of check most crash write-ups skip. Three things still need work. First, the claim that damage "tracked one thing, how much of each index was a bet on AI hardware" has a weight figure only for the STAR 50 (over 83% semiconductors, a single Huxiu…
The warning signs were about price paid and crowding, not a single headline. Press reports put the STAR 50's trailing P/E near 100x, against a peak near 250x and roughly 30x for the Nasdaq 100 (Reuters). The index also took on more of the theme on the way up: the reconstitution effective after the 2026-06-12 close added Hua Hong, Yuanjie, Moore Threads and MetaX, so passive holders were buying more AI-chip exposure just before the peak (the ETF 588000.SS closed at 2.344 on 2026-06-30). Trading was already feverish then: exchange data show daily turnover of roughly 160 to 235 billion yuan in…