CITIC Securities: A-Share Correction Nearing Its End, AI Core Stocks' Decline Now Rivals New Energy's 2022 Selloff
nashnova research
CITIC Securities says AI core hardware stocks have dropped 41.1% from their peak, nearly matching the 42.7% first-round correction in new-energy leaders in 2022; institutional selling pressure is fading, but a confirmed reversal signal has yet to appear.
Who cracked first in this sell-off?
Institutional heavy-hitters led the fall: the top-20% basket by institutional ownership has dropped 13.6% since September 22, versus 6.8% for the bottom-20% basket. This means → the selling started with fund managers cutting positions, not retail panic.
Margin-financed names followed: the highest-leverage group fell 11.7%, the lowest just 5.9%. In plain terms = institutions dumped first, then leveraged money was forced to follow — a two-wave chain reaction.
The hardest-hit sector is AI hardware (electronics, telecom equipment), where the size of each stock's decline tracks its institutional ownership almost one-to-one.
How far has this correction gone?
The "AI Core 15" — the 15 AI hardware stocks most heavily held by active equity funds at mid-year — have fallen 41.1% from their July 1 peak. In 2022, the "New Energy Core 15" bottomed out at a 42.7% drawdown. The two rounds are now nearly identical.
At the index level, electronics is down 41.2% and telecom 42.9% from their respective highs, close to the 43.5% first-round drop in the power-equipment and new-energy index last cycle. This reflects a repeating pattern: when a super-cycle's "second derivative" — its acceleration — downshifts, hardware valuations correct by a historically comparable margin.
Yet the all-clear has not sounded: the 10-day rolling return spread for institutional heavyweights is still -7.9%, and within electronics and telecom the gap is -11.6 percentage points. In plain terms = institutional money is still flowing out; the turning point is not here yet.
Are there any early signs of a bottom?
Some late-stage exhaustion signals have appeared: the "daily price move / turnover" ratio for the top-200 institutional holdings has jumped to its year-high. This means → the same trading volume is now moving prices more — a sign that loose shares are being flushed out.
Stock-level divergence widened noticeably on Thursday and Friday; indiscriminate position-cutting is easing — it is no longer "sell everything."
Broad-market ETFs saw heavy inflows: over October 15–16, the CSI 300, SSE 50, ChiNext, and STAR 50 ETFs drew a combined net inflow of ¥47.7 billion. Meanwhile, government-bond ETFs attracted ¥10.2 billion in net subscriptions over the past 10 trading days — the 99th percentile since 2023. This reflects a split mindset: funds are bottom-fishing equities and buying bonds for safety at the same time.
What is really driving this correction?
On the surface it is about stretched valuations. Underneath, it is a dual squeeze from weaker AI monetization expectations and high interest rates. Recent ARR (annualized recurring revenue) growth at OpenAI and Anthropic came in below expectations, deepening doubts about whether existing AI use cases can generate enough new cash flow to justify compute spending.
The funding-side pressure is tangible: this year the five major cloud operators have issued ¥228.8 billion in bonds, and emerging AI-infrastructure players another ¥134.1 billion. The combined ¥362.9 billion of net new issuance is roughly 40% of the U.S. Treasury's total issuance of 10-year-plus bonds over the past year. In plain terms = the AI arms race is absorbing a huge share of the bond market's capacity, pushing rates higher across the board.
CITIC Securities adds a caveat: for long-term capital focused on AI growth's "first derivative" — the growth rate itself — rather than its "second derivative" — the acceleration — this correction does not break the trend. Memory prices, compute-rental rates, and core cloud operators' cloud-segment margins have not hit an inflection point.
How do interest rates and AI investment connect?
CITIC Securities frames high rates and AI investment as two sides of one coin. Lower AI investment expectations → less expected bond supply → rates may actually fall. Recently, as North American AI hardware stocks corrected, long-end U.S. Treasury yields also began a sustained decline — validating exactly this logic.
This means → if the market keeps marking down AI investment expectations, sectors that have already corrected deeply outside AI may offer new structural opportunities.
If AI spending slows, expectations for domestic-demand policy stimulus may heat up in turn. Put simply = extrapolating the negative outlook in a straight line risks missing a structural rotation from AI into non-AI sectors.
市场有风险，内容仅供研究参考，不构成投资建议。