Six Historical Rules of A-Shares Have Broken Down — AI Remains the Core Theme in H2
Claire Weston
GF Securities' strategy team finds that six long-standing A-share rules validated over two decades have all failed in this cycle, driven by the end of the property super-cycle and the peaking of demographic dividends — the earnings gap between AI and non-AI sectors is likely to widen further in H2, leaving little fundamental support for a style rotation.
Which six "iron rules" have broken?
A-shares had never re-rated for three consecutive years — but 2024–2026 marks the third straight year of valuation expansion. Electronics and telecom have led the market into a fourth year, breaking the rule that sectors rarely outperform for more than three.
Institutional holdings in electronics have exceeded 20% for over five quarters, yet the sector keeps rallying — historically, hitting 20% was almost a reliable top signal.
Telecom was the most heavily added sector for four consecutive quarters; subsequent excess returns were +31%, +14%, +10%, +37% — destroying the old rule that "the most aggressively bought sector always underperforms next quarter."
TMT turnover share has breached 45%, above the prior assumed ceiling of 40%. Meanwhile, finished-goods inventory bottomed in June 2023, yet three years later no large-scale fiscal stimulus has materialized. This means → after the property cycle's exit and the demographic peak, A-share profit and market-cap structures have fundamentally shifted, rendering the old playbook obsolete.
Which rules are approaching a tipping point?
As of June 24, the top-5% stocks' share of turnover (20-day moving average) reached 49.8%, closing in on the 2015 and 2021 historical highs. In plain terms = market money is concentrating into fewer and fewer names, nearing an extreme.
Drawing on the U.S. dot-com cycle, GF argues this ceiling could be breached — if it is, the trend of "a handful of stocks absorbing most liquidity" can run further.
Sector valuation dispersion — measured by the standard deviation of industry PB historical percentiles — has hit an all-time high. But GF reviewed six prior episodes of extreme dispersion and concluded: peak valuation divergence does not equal a bull-market top.
What is the K-shaped split, and will it close in H2?
H1 A-share pricing showed a clear K-shape: companies in the top 10% by revenue growth averaged over 40% gains year-to-date, while stocks with strong cash flow, high dividends, and low valuations actually fell harder. This reflects the global economy's own K-shaped divergence — regions with AI supply-chain assets are booming; those without are weakening.
The upper arm: the global top-five cloud companies plan $769 billion in 2026 capex, revised up again from the start of the year. H100 rental prices topped $2.80/hour. Doubao's daily token calls broke 180 trillion in June; Google's monthly token consumption approached 3,000 trillion.
The lower arm: China's fiscal stance remains restrained, cushioned by resilient external demand. This means → whether domestic policy truly ramps up in H2 hinges on whether the overseas restocking impulse fades. Bank net interest margins have dropped to a record low of 1.40%, making a broad rate cut unlikely. Fiscal spending ran at −0.3% YoY through January–May; Q3 may see modest backfilling, but the bias stays cautious.
Where is the money coming from — and going — in H2?
H1 saw roughly ¥860 billion in net retail inflows, up from about ¥690 billion in H2 2025. But H2 faces headwinds: margin-trading momentum is slowing, northbound flows are capped by U.S. Treasury yields, industrial shareholders are accelerating sales around a lock-up expiry peak, and mega-IPOs such as CXMT are expected to divert funds. The net supply-demand gap is projected to narrow by over ¥100 billion.
Retail still has room: demand-deposit deviation from trend is about ¥1 trillion short of prior "great migration" peaks. The share of active equity funds back to breakeven has reached 50.8%, significantly easing redemption pressure.
Institutions are taking profits: active equity-mix fund positions have dropped to 85.7%, and active private-fund positions have slipped to 62.3%. Insurance capital is the key stabilizer — Q1 2026 insurance inflows hit the highest level since 2013, and further incremental buying is expected in H2. In plain terms = retail investors still have cash to move in, but institutions are selling — the market is shifting from an inflow-driven phase to a turnover-driven one.
Beyond AI, which sectors have fundamental support?
Energy-storage lithium batteries are seen as the most certain non-AI sector reaching a recovery inflection: China's June lithium-cell production schedule hit roughly 268 GWh, a record high for the fourth consecutive month. A recent pullback in lithium-carbonate prices eases pressure on storage-project IRR — the internal rate of return that determines whether a project is profitable.
Non-ferrous metals: gold's structural thesis is intact — rising global sovereign debt, a weakening dollar reserve status, and central-bank buying. Copper's supply-demand gap is even more structural: global copper output turned negative YoY starting March 2026.
Innovative-pharma leaders trade below their mean minus one standard deviation, but need two triggers to re-rate: easing U.S.–Iran tensions and a leg down in U.S. Treasury yields. This means → these sectors have fundamental floors, but timing depends on external variables — not on rotational flows out of AI.
What is the single most important takeaway for H2?
GF emphasizes that amid all the broken rules, one first principle still holds: "relative earnings advantage determines relative stock-price performance."
AI-sector EPS keeps getting revised upward; non-AI earnings show no sign of accelerating — the earnings-growth gap between AI and non-AI will most likely continue widening in H2.
In plain terms = for other sectors to see a genuine broadening of gains, their own fundamental inflection must actually arrive — it won't come from money "spilling over" from AI. Until that inflection materializes, the AI theme's relative edge is hard to shake.
Content is for reference only, not financial advice.