Zhang Yu, Chief Economist at Huachuang Securities, is known for her unique, forward-looking, and logical insights. In a recent online interview, she shared her latest views on the future of AI tech trends, the global economic landscape, manufacturing development, consumption prospects, household wealth migration, and gold allocation.
Key takeaways: 1. If there is no trend reversal, the tech rally is not over. 2. Historically, new general-purpose technologies like electricity and the internet initially see labor productivity decline before accelerating. 3. The early winners of the AI industry may not be the final winners. 4. There are two fundamental differences between the current AI wave and the 2000 dot-com bubble, suggesting a structural, differentiated rally. 5. In 2000, the US controlled the entire profit chain of the internet; today, AI hardware and component profits are distributed across different economies. 6. Globally, among the top 50 most frequently used models, China accounts for about 20, and the US about 30, challenging existing valuation frameworks. 7. Gold returns often come in short, explosive pulses, making it a difficult asset to hold.
Don't Mechanically Apply Historical Patterns
Zhang Yu cautioned against mechanically applying historical comparisons. For example, recent fluctuations in free cash flow for companies like Amazon and Google don't necessarily signal a shift in industry outlook. She believes that predicting the next phase of tech requires a deep understanding of cutting-edge research and industry progress. “If there is no trend reversal, the tech rally may not be over. If there is, the inflection point may appear after a period of consolidation.”
AI Technology Will Experience a Productivity 'J-Curve'
Zhang Yu views AI as a general-purpose technology, not a specialized one. Historical data on electricity and the internet shows that productivity improvement follows a 'J-curve,' often declining before accelerating. She attributes this initial decline to three factors: existing production relationships and organizational structures may not fit new technology (like running a horse cart on rails); statistical accounting lags behind technological change; and workforce retraining requires investment with no immediate efficiency gains.
Early Winners Are Usually Not the Final Winners
Capital markets undergo a similar trial-and-error process. The initial demand often leads to massive hardware investment, but key factors are expensive, and early attempts may fail. Only after supply expands and prices drop does mass adoption occur. Zhang Yu points to mobile internet: before 'speed upgrades and fee reductions' around 2013-2014, data costs were high, limiting app penetration. Applications emerge organically after factor prices fall. She believes the same applies to AI: hardware prices are still too high, and true mass-market applications will only appear after supply increases and prices fall. “Only then will the real big winners emerge.”
Two Fundamental Differences from the Dot-Com Bubble
Zhang Yu identifies two key distinctions between the current AI wave and the 2000 bubble, suggesting a structural, differentiated market rather than a synchronized global peak. First, in 2000, the US controlled the entire profit chain. Today, AI hardware and component profits are globally distributed. Second, in 2000, quality tech companies were concentrated in US listings. Now, competitive models and companies exist outside the US, like the top 50 models where China accounts for 20 and the US 30, challenging existing valuation systems.
AI is Attractive, but Chinese Manufacturing is More Than Just AI
Discussing A-share volatility and China's manufacturing resilience since July, Zhang Yu describes the past two years as 'very turbulent.' Despite tariff shocks, China's export share has risen. Amid cost pressures from rising metals prices and supply chain disruptions from geopolitical events, Chinese manufacturers have maintained stable deliveries. She notes that high energy costs and disruptions have forced some European and Southeast Asian capacity out, with irreversible exits. In contrast, Chinese manufacturing has proven highly resilient, leveraging its complete industrial system, stable delivery, and supply chain agility. However, she believes capital markets are overly focused on AI within manufacturing. “AI is the most attractive, but Chinese manufacturing has many beautiful things.” She notes that heavy trucks, machinery, machine tools, shipping, and shipbuilding have also shown strong exports, performance, and orders. AI is just the fastest 'advance guard' of the broader manufacturing rise.
High Capex Is Not a Protective Barrier for AI Stocks
Addressing the high capital expenditure growth in AI, Zhang Yu offers a counterintuitive conclusion: high Capex is an industry indicator, not a leading indicator of stock price safety. Her team analyzed 18 financial metrics from 7 iconic companies during the 2000 bubble. The six effective indicators were: operating cash flow (OCF) YoY, OCF/Capex, revenue YoY, absolute free cash flow (FCF), absolute EBITDA, and Capex YoY growth. Capex YoY was a lagging indicator, while the others were leading. During the bubble, 5 of 7 companies saw their stock prices peak while Capex was still growing. “Many people emphasize 'high Capex' when discussing AI stocks,” she says. “But high Capex is an industry issue, not a stable leading relationship with stock prices. It's actually a lagging indicator.”
Operating Performance Is the Stock Price's 'Gatekeeper'
In contrast, absolute FCF and EBITDA showed more stable leading significance. Zhang Yu calls them the 'gatekeepers' of stock prices. In the historical analysis, each had an effective accuracy of about 5/7. For roughly 5 companies, when FCF or EBITDA peaked, the stock price turned within the same quarter (0-3 months). “FCF and EBITDA are the true gatekeepers,” she says. “If they turn down, many trends may have serious flaws, and the stock price will follow.” Other indicators have some leading value but with unstable lead times, making them better for cross-validation than as standalone signals.
Will the Consumption Sector See a Rotation?
Regarding a potential rotation to consumption, Zhang Yu highlights the link between income divergence and industrial restructuring. Using listed companies as a sample, the top 10% of profitable manufacturing firms may capture 90% of industry profits but employ only 45-50% of the workforce. “If 100 people work in manufacturing, about 45 share 90 yuan of profit, and the remaining 55 share 10 yuan.” High-income groups have a lower marginal propensity to consume, while lower-income groups have limited consumption potential. As manufacturing shifts to capital-intensive, automated models, profits and income naturally concentrate. Therefore, without changing primary distribution, policies are more likely to adjust secondary distribution. She believes structural policies are more effective than aggregate ones in a K-shaped environment. For consumption subsidies, she recommends continuing durable goods subsidies, which should not be seen as simply borrowing future consumption but as smoothing the release of consumption potential amid industrial upgrades and income divergence. She suggests expanding the scope to include services.
Gold's Decade-Long Bull Thesis Unchanged
Despite recent gold volatility, Zhang Yu maintains her 'decade-long bullish' view on gold. She says the three underlying layers of logic supporting this long-term thesis are strengthening, with no clear changes. Gold's low correlation with other assets provides value in portfolio allocation. However, she suggests a 5% allocation in household investable assets is appropriate. “If you can hold 5%, it's fine.” She believes short-term gold trading is extremely difficult. Unlike famous investors in bonds, tech, or commodities, there are few 'gold gods' because the spot market is dominated by sovereign reserve managers, making it hard for individuals to predict. Gold's returns also differ from stocks or bonds. It can be flat for long periods, then surge briefly. “You might not make money for 4 years and 280 days, then gain 50% in the last 80 days, or even the last 10 days.” This means investors can't easily time the key rallies. For regular investors, frequent timing is not more effective than a reasonable allocation. Zhang Yu repeatedly warns that gold has poor holding experience, no interest, high volatility, and a low Sharpe ratio, making it unsuitable for large bets or essential funds. The longer the investment horizon, the higher the probability of realizing returns.
If AI is Not Disconfirmed, Continue 'Separating Wheat from Chaff'; If Disconfirmed, Seek 'Two-End Benefits, Irrelevant Middle'
For the H2 capital market, Zhang Yu presents two scenarios. First, if the AI tech trend is not disconfirmed, the market may follow the current trend, but the tech sector will further 'separate wheat from chaff,' with capital favoring assets with real, quality fundamentals. New sub-sectors may also lead. With domestic economic transition continuing, she suggests a relatively positive stance on equities, while bonds and FX should be neutral to cautious. Second, if the global AI trend (led by the US) is disconfirmed, the US market might shift to Treasuries and domestic consumption. Meanwhile, mid-stream manufacturing outside AI could gain more attention. Her basic strategy is to seek 'two-end benefits, irrelevant middle.' However, she reiterates that tech assets are volatile, and judgments depend heavily on industry foresight. For macro researchers without information advantages, the focus should be on data, reports, and financial indicators, not static narratives.
Comments