Goldman Sachs Identifies 10 Signals to Gauge AI Sentiment in China

China's AI hard tech sector, after a rapid rise in the first half of the year, is entering a phase of emotional reassessment.

According to Goldman Sachs' China Strategy report released on August 9, AI hardware stocks in its defined China AI stock universe rose an average of 33% in the first half of the year. As global AI hardware trading reverses, the STAR 50, ChiNext Index, and CSI 1000 have all declined more than 20% from their recent highs.

The question is: Is this pullback merely a short-term profit-taking, or does it signal a more profound shift in AI trading?

Goldman Sachs did not merely look at the index decline but assessed the sentiment and correction cycle of China’s AI hard tech across ten dimensions: earnings dispersion, market concentration, valuation, leverage, retail investor sentiment, institutional positioning, corporate behavior, policy, state-backed capital, and earnings forecasts.

Ten signals together point to a moderate conclusion: much of the prior speculative positioning, overvaluation, and leverage have been unwound, but concentration risk remains, and fundamental upward momentum is slowing. The market’s most crowded phase may be behind us, but it’s too early to confirm that the correction has fully ended.

In the first half of the year, the defining characteristic of China's AI trading was not a general rise in all AI assets, but rather heavy capital concentration in hard technology.

According to Goldman Sachs data, the return differential between the SSE Science and Technology 50 Index and the Hang Seng Tech Index widened by more than 100 percentage points in the first six months of this year, reaching levels comparable to the peak dispersion seen in early 2021. Small-cap, high-growth, and momentum styles also emerged as the primary drivers of market performance at that time.

After entering July, these trends reversed rapidly. As global AI hardware trading cooled down, the excess returns of hard tech, small-cap, and momentum factors significantly retreated, and the performance gap between soft tech and hard tech has returned to near its long-term average.

The adjustments to momentum and size factors have even entered more extreme ranges. Based on this, Goldman Sachs concludes that, from a price momentum perspective, the most rapid phase of drawdown and style reversal may begin to moderate.

This signal is generally positive: the sectors with the largest prior gains and the most crowded positions have already experienced significant cooling. However, a slowdown in the pace of price declines does not mean that the related stocks have fully been cleared out.

The second signal comes from market breadth.

Goldman Sachs estimates that 90% of the Sci-Tech Innovation Board's gain this year has been driven by the top 10 performing stocks. For comparison, this proportion was 23% at the beginning of 2021 and 46% in mid-2015 for the ChiNext Index.

Trading volume is also becoming concentrated in a few sectors. The technology sector, ChiNext, and the Sci-Tech Innovation 50 currently account for 26%, 14%, and 6% of A-share cash trading volume, respectively, all at relatively high levels in recent years.

Meanwhile, the correlation among individual A-share stocks remains at a recent low. Low correlation indicates that the market is not primarily trading on broad macroeconomic factors, but rather continues to focus on a few micro themes, such as AI and technology.

This indicates that although hard tech stock prices have corrected, AI remains the market’s most important pricing driver, with earnings and trading activity yet to truly broaden. Sentiment has cooled somewhat, but trading concentration remains unchanged.

Valuation is the third signal for determining whether sentiment has been fully released.

At the June high, the forward P/E ratio of the ChiNext Index, weighted by market capitalization, was approximately 26x, while the median forward P/E ratio of the STAR 50 components was around 50x. Following the subsequent decline in stock prices, the valuation multiples of A-share hard tech companies have now fallen to or below their long-term average levels.

When factoring in earnings growth, the forward PEG ratios of relevant companies are also at or below historical medians, indicating that some of the previous overheating has been absorbed when viewing valuations relative to their historical levels.

However, the absolute valuation of the Sci-Tech Innovation 50 remains relatively high. After a synchronized correction in the global AI sector, the valuation discount of Chinese AI-related stocks compared to their overseas peers has significantly narrowed, thinning the previous safety margin that "AI assets in China were cheaper."

Therefore, the valuation signal does not indicate that things are "already cheap," but rather that the most expensive phase may be over. The market will now rely more on earnings realization rather than continued valuation expansion.

The margin balance is the most direct indicator for observing leveraged trading in the A-share market.

According to Goldman Sachs data, A-share margin balances have declined from approximately RMB 3 trillion to RMB 2.6 trillion, and the ratio to free-float market capitalization has fallen from 6.0% to 5.5%, indicating that some leveraged funds have exited the market.

However, both margin balances and the margin ratio remain above historical norms. Goldman Sachs believes that, compared to the more pronounced deleveraging recently seen in the South Korean and Taiwanese markets, A-share deleveraging may still be in an earlier stage if deleveraging continues.

Particular attention should be paid to the distribution of leverage. According to Goldman Sachs’ calculations, the top 10% of stocks with the highest margin balances currently account for approximately 30% of all margin financing in the A-share market, a record high, and are concentrated primarily in the AI and hard technology sectors.

This means that overall market leverage risk may be significantly lower than in 2015, but structural pressures still exist within the AI and hard tech sectors. If related stocks continue to decline, concentrated margin positions could still amplify volatility.

Goldman Sachs's fifth signal comes from retail investor sentiment.

Although the proportion of domestic public funds, pension funds, and insurance capital continues to rise, retail investors still account for approximately 70% of A-share daily trading volume. As a result, changes in retail investor sentiment continue to directly impact short-term market volatility.

Goldman Sachs' retail investor sentiment barometer includes high-frequency metrics such as margin data, new account openings, IPO subscriptions, turnover rates, and stock allocations. The indicator has now declined from about one standard deviation above the one-year average a month ago to near zero standard deviations.

This means that retail investors' risk appetite in the A-share market has cooled from a relatively warm level to neutral, or even slightly subdued.

Retail sentiment has cooled rapidly, indicating that some speculative pressure has been released in this correction. However, a zero standard deviation does not signify extreme pessimism or a traditional "panic bottom." It merely suggests that the momentum of buying on rallies has faded, but it does not yet prove that the market has completed its final round of selling.

Institutional investors' behavior is more complex than that of retail investors.

According to Goldman Sachs data, the total assets under management by domestic public mutual funds have approached 4 trillion RMB, with approximately 700 billion RMB allocated to equities, accounting for 6.6% and 15% of the total market capitalization and free-float market capitalization of A-shares, respectively.

During the market correction, the cash ratio of equity mutual funds has increased, indicating that fund managers have moderately reduced risk. However, their allocation and overweighting of technology stocks such as semiconductors, hardware, and software remain at historical highs.

The risk reduction from systematic strategies may be more pronounced. The decline in cash trading volume and financing spreads for small- and mid-cap stocks indicates reduced activity among systematic investors, such as quantitative funds, suggesting they may have undergone deeper deleveraging.

There is clear divergence within this signal: quantitative and short-term capital has contracted, but traditional institutional holdings in core tech stocks have not significantly weakened. As long as institutional allocations remain high, AI hard tech cannot be defined as a fully decongested trade.

A company's own actions often better reflect insiders' judgments on valuation and risk than market slogans.

Goldman Sachs examines corporate behavior from three perspectives: repurchases, trading alerts, and significant shareholder transactions. On a quarterly basis, A-share repurchases in the third quarter of 2026 have reached a multi-year high, with the number and value of announced repurchase transactions increasing by 35% and 59% year-over-year, respectively.

An increase in buybacks typically indicates that management believes the company's stock is undervalued relative to its intrinsic worth, or at least that they are willing to deploy cash at current prices to support shareholder returns.

Another change is that announcements regarding abnormal price fluctuations and trading risks by listed companies increased noticeably in June, prior to the market correction, but have since declined significantly. Trading by major shareholders and company executives has also shifted from net selling for most of the first half of the year to a more balanced state.

This set of signals is overall positive: corporate insiders' behavior is no longer trending toward selling and signaling overheating as in the later stages of a rally, but is gradually shifting toward buybacks and reduced net selling.

Sentiment cycles in China's stock market are often influenced by policy changes.

The deleveraging in 2015 and the regulatory tightening that began at the end of 2020 both served as key catalysts for market reversals; conversely, clear policy support has multiple times driven strong rebounds in Chinese equities.

Goldman Sachs uses large language models to analyze public statements from regulators and policymakers, measuring the risks of policy support and tightening in the stock market based on the frequency and intensity of wording.

The model shows that concerns by policymakers about market overheating and the resulting risk of policy tightening peaked in the first quarter of 2026 and have since declined to a more neutral range.

This means that policy is currently neither a clear emotional catalyst nor the primary source of pressure in this adjustment. Compared to price, leverage, and profit factors, policy signals are temporarily closer to neutral.

The ninth signal comes from the "national team."

Goldman Sachs estimates that the broad "national team" currently holds approximately RMB 5 trillion in A-share assets, accounting for 5% of the total market capitalization of A shares. Historically, this type of capital has tended to buy against market pressure and may sell at higher levels when markets rise and valuation attractiveness declines.

According to Goldman Sachs' tracking data, after selling approximately RMB 1.5 trillion in A-shares over the previous six months, the "national team" has shifted to a net purchase of over RMB 140 billion in the past three weeks, including a small amount of Sci-Tech Innovation 50 ETFs.

The "national team" has shifted from selling to buying, indicating that policy-related funds have changed their assessment of market risk and have provided some downward support for the broader market.

However, this alone cannot confirm a mid-term bottom. Goldman Sachs’ historical backtesting shows that a mid-term market bottom is more likely only when the state-backed investors’ weekly net purchases exceed 1.5 standard deviations. The clearer conclusion at present is that support has reemerged, not that the market has received an unconditional floor.

The final and most decisive signal for how far AI's market momentum can go comes from capital expenditures and profit forecasts.

Goldman Sachs estimates that this year, the nine major hyperscale companies and cloud service providers in the United States and China may spend over $900 billion on AI, increasing further to $1.3 trillion next year, equivalent to 1.7% and 2.3% of the combined GDP of the U.S. and China, respectively.

These expenses ripple through the global AI supply chain, impacting South Korean memory chips, Taiwanese wafer foundries, Japanese semiconductor materials and equipment, as well as China’s power, infrastructure, and technology companies.

This year, the capital expenditure forecasts for eight U.S.-listed mega-cap companies for fiscal years 2026 and 2027 were raised by 32% and 75%, respectively, driving upward revisions of 12% and 22% to the profit forecasts for China’s hard tech sector for fiscal years 2026 and 2027.

This indicates that the rise in China's AI hard tech is not merely driven by sentiment and leverage, but is genuinely supported by increased capital expenditure and profit growth.

The issue is that the pace of upward revisions to capital expenditure forecasts has slowed from its peak, and the momentum for upward revisions to earnings forecasts for China’s hard tech companies shows at least temporary signs of peaking. In contrast, soft tech still has room for cyclical improvement in earnings revision momentum.

Therefore, in the coming months, the key factor influencing AI sentiment will no longer be just the scale of capital expenditures, but whether these expenditures continue to exceed expectations and whether cloud providers can offer clearer paths to AI monetization.

Putting together ten signals, the current position of China’s AI hard tech becomes increasingly clear.

Positive signals include: the extreme performance gap between hard and soft technologies has narrowed, momentum and small-cap styles have rapidly reversed, valuations have returned to near long-term averages, retail investor sentiment has turned neutral, corporate buybacks have increased, major shareholder selling has balanced out, and the "national team" has resumed net buying.

These changes indicate that a significant portion of the speculative positions, overvalued valuations, and leverage pressures accumulated in the first half of the year have been released.

However, cautious signals also persist: the absolute valuation of the Sci-Tech Innovation 50 remains high; trading and earnings concentration in China’s tech stocks are at elevated levels; margin balances still exceed historical norms; leverage concentration has reached a new high; and mutual funds’ allocation to tech stocks remains at historical highs.

Fundamentals lie between the two. Global AI capital expenditures remain strong, and earnings forecasts for hard tech are being revised upward; however, the momentum behind these upward revisions in capital expenditures and earnings forecasts has begun to slow.

Goldman Sachs still maintains an overweight stance on A-shares and remains fundamentally bullish on AI hard technology in the long term, but emphasizes rotation and diversification in the short term. Its recommendations include gradually increasing allocations to selected Hong Kong soft technology stocks, policy beneficiaries, and self-reliant sectors, while also focusing on stocks with upward earnings revisions, IPOs, and cash returns from dividends and buybacks.

For investors, these ten signals do not simply indicate an "AI bottom" or the "end of the AI rally," but rather a sentiment thermometer: the hottest phase may already be behind us, the steepest declines might be starting to slow, but position concentration and profit expectations still need to be fully absorbed.

In the next phase, whether China’s AI trading can reignite will depend more on corporate profitability and AI monetization than on renewed reliance on valuations, leverage, and FOMO.

律动 BlockBeats