China's top AI hedge funds have started booking profits, with Shanghai Everlead Capital, up 164% this year, leading the funds now trimming their biggest winners. As per BeInCrypto's exclusive layer data, money is rotating out of the hottest AI trades, with Everlead trimming optical and chip-packaging stocks that had gone vertical. The fund sold because names like Zhongji Innolight's trillion-yuan market cap and Yangtze Optical Fibre's twelvefold rally show how far the AI optical trade ran, with gains that size inviting profit booking. Hunjin Capital moved similarly, trimming most crowded AI holdings and rotating into cheaper traditional stocks, by its own measure indicating the AI hardware cycle is now 60% complete, double its February reading. This represents the first concrete evidence of institutional investors taking profits after months of exceptional performance.
The profit-taking extends beyond individual funds to the entire AI ecosystem, with compute stocks gaining about 62% but falling roughly 13% last month, while power and infrastructure rose about 11%, then stalled. Apps and software are the opposite, lagging all year down about 9%, then gaining roughly 5% last month as fresh money moved in. A proprietary gauge tracks the 30-day correlation between power and compute baskets at 0.74, up from near neutral earlier in the cycle, showing how AI's bottleneck has moved from chips to electricity with data-center power demand set to roughly double by 2030. This looks like a late-cycle rotation, not a collapse, with the real question whether the whole market is turning with the funds as they book profits.
The extraordinary rally in artificial intelligence chipmakers is showing signs of losing momentum as investors reassess lofty valuations and question whether the rapid pace of spending on AI infrastructure can be sustained. The Philadelphia Semiconductor Index has more than doubled over the past 12 months, even after falling nearly 18% from its June peak, while the equal-weighted S&P 500 has gained about 11% and Europe's STOXX 600 has advanced around 8%. Bank of America's July global fund manager survey showed that 82% of respondents considered semiconductors the most crowded trade in global markets, while none reported holding bearish positions against the sector. However, expectations are beginning to shift as UBS estimates project hyperscaler capital expenditure to jump 76% this year to $673 billion, but growth is expected to slow sharply to 25% in 2027 and just 6% in 2028.
The entire AI ecosystem's future hinges on 2027 capex commitments, with big cloud firms forecasted to commit more than $600 billion to the buildout in 2026, up about 36%, and forecasts pushing it past $1 trillion in 2027. The threat is a 2027 plateau, and a price war could force one, with Chinese models now matching top US systems at a fraction of the cost, some about 55 times cheaper. If that return breaks, big tech cuts capex, the compute and power layers fade for good, and the funds' early profit-taking becomes the first sign of a bubble burst. The bulls still see real profits, not a 2000-style bubble, while the bears see the price war breaking those returns first. The biggest AI companies continue generating the strongest earnings growth, with Nvidia, Microsoft, Meta Platforms, Amazon, and Broadcom still investing tens of billions of dollars into AI infrastructure, suggesting the bull market may have more room to run despite narrowing leadership.
Financing AI expansion faces greater scrutiny with softening bond demand, as corporate bond issuance by Big Tech has reached billions of dollars this year. Apollo's analysis shows bond cover ratios have fallen below two times in July from nearly five times in February, indicating weakening investor demand relative to supply. The Bank for International Settlements warned in June that weaker-than-expected returns from AI investments could eventually reduce financing availability and turn today's spending boom into a prolonged downturn. Beyond financial concerns, AI infrastructure expansion is encountering growing political and community resistance, with opposition to large-scale data center developments increasing across the United States due to concerns over electricity consumption and infrastructure impacts. New York became the first U.S. state to impose a one-year moratorium on large new data center construction, reflecting mounting concerns over the environmental and infrastructure costs associated with the AI boom.
