WE Think: Super Cycle or Speculative Excess?

I recently published a book, Long-Term Wins: Navigating Bull and Bear Cycles in the Stock Market, with the help of my colleagues and the editorial team at CITIC Press. It draws on 33 years of investment experience, pulling together material from our Ruisen Academy training sessions and nearly two decades of Weizhi Insights essays. I hope readers will find it a useful reference for thinking through market cycles.

AI Has Become the Market's Primary Driver

The first half of 2026 saw rising equity markets, but beneath the surface, performance varied wildly. The single most important factor separating winners from losers was exposure to AI.

South Korea's KOSPI surged 101.14%, Taiwan's Weighted Index climbed 59.25%, and the Nasdaq gained 12.79%. In mainland China, the ChiNext and STAR Market rose 35.58% and 53.99% respectively—far outpacing the Shanghai Composite and the CSI 300. What did the best-performing markets have in common? They all sit on the AI value chain, especially in semiconductor manufacturing and AI infrastructure.

The launch of ChatGPT accelerated a global race to build large language models and the infrastructure required to support them. Google, Amazon, Microsoft, and Meta are projected to boost their combined capex from roughly US$150 billion in 2023 to US$720 billion by 2026. China's top internet firms are following a similar playbook.

This spending surge has strained supply across the hardware value chain—from memory chips and semiconductor fabrication to optical modules, fiber optics, electronic fabrics, MLCCs, and even power infrastructure. In many segments, supply simply hasn't kept up, pushing prices higher, lifting earnings, and triggering fresh waves of capacity expansion.

A Technology Investment Cycle Unlike Any Other

From an industry perspective, the scale of the current investment cycle is difficult to overstate.

Consider the scale: across 2025 and 2026, the five largest U.S. cloud providers are expected to invest roughly US$1.2 trillion in combined capex. China's top seven tech firms will add another RMB1.2 trillion. For some context, total U.S. residential construction spending—including renovations—is projected at around US$1.4 trillion for 2026 alone. AI data-center investment alone accounts for over US$300 billion of that incremental outlay.

Importantly, AI adoption remains in its early stages across much of the global economy. The infrastructure buildout currently taking place is occurring before most industries have fully integrated AI into their business models, suggesting substantial runway for future demand.

The investment footprint stretches across a remarkably broad supply chain. Building a data center is like assembling a giant bucket—GPU, HBM, optical transceivers, fiber, liquid-cooling systems, PCBs, MLCCs, power infrastructure. Every gigawatt of capacity requires roughly US$43 billion in investment. Given the economics of AI computing, operators can tolerate much higher component costs than traditional consumer-electronics manufacturers, creating unusual pricing power for suppliers when bottlenecks emerge.

As a result, prices have risen sharply across memory, optical fiber, electronic fabrics, semiconductor materials, and MLCCs. Companies in these segments have benefited from simultaneous volume growth and pricing expansion, leading to significant earnings acceleration and strong equity performance.

Although supply expansion is underway, meaningful relief is unlikely to arrive immediately. Samsung and SK Hynix recently announced plans to invest approximately US$874 billion over the next decade and double production capacity within five years. However, the majority of additional capacity is expected to come online after 2028, leaving supply conditions relatively tight through 2026 and 2027.

Based on our industry research and recent company meetings, conditions across China's AI hardware ecosystem remain exceptionally strong. Earnings guidance from storage and optical-fiber companies continues to exceed expectations. In more than three decades of investing, we have not witnessed a technology-driven investment cycle of comparable scale. From both a capital-spending and industrial perspective, the current AI buildout stands apart from previous technology upgrade cycles. Its economic significance may ultimately exceed China's 2008 stimulus program and rival or surpass the impact of any major U.S. housing cycle in recent decades.

The Investment Question: Does Strong Growth Guarantee Strong Returns?

While the industrial case for AI appears compelling, investing is a different exercise altogether.

Howard Marks draws a useful distinction between first-level and second-level thinking. The first-level mind looks at a fast-growing industry and concludes it must be a good investment. The second-level mind pauses and asks a harder question: has the market already priced that growth in?

For investors, the most important question is not whether AI is experiencing a boom. It clearly is. The real question is whether current valuations are underpricing, fairly pricing, or overpricing future growth.

The 2021 case of China's premium liquor stocks is instructive. Investors who bought Kweichow Moutai and Shanxi Fenjiu at peak valuations endured years of negative returns. Yet the companies themselves remained excellent—their competitive positions didn't deteriorate. The problem wasn't quality; it was price. Too much future growth had already been baked into the share price.

The same risk exists in parts of today's AI supply chain.

Investors must look beyond current earnings growth and examine industry structure, competitive dynamics, and barriers to entry. In segments where capacity can be expanded relatively quickly and competitive advantages are limited, cyclical forces often dominate structural growth trends. Extrapolating today's unusually high margins, returns on equity, and earnings growth far into the future can be dangerous.

Within our own investment review process, some of our largest mistakes occurred in cyclical-growth sectors where we underestimated the impact of future capacity expansion. As industry conditions normalized, earnings and valuations declined simultaneously, creating severe losses. Time often becomes the enemy of highly profitable industries.

Of course, long-term investors and short-term traders may arrive at entirely different conclusions. The key is not to imitate someone else's strategy, but to remain consistent with one's own investment framework and time horizon.

The Challenge of Identifying the Turning Point

Many AI-related stocks have delivered extraordinary returns, yet identifying the turning point of a major investment cycle remains exceptionally difficult.

George Soros once observed that successful investing requires spotting the prevailing narrative, joining before it becomes consensus, and having the discipline to step off before that consensus turns to euphoria. His famous dictum applies here: buy when others are afraid to buy, and sell when others feel they can't afford not to.

This raises an important question for today's market.

For AI investors, the key issue may no longer be whether the industry opportunity exists. Instead, the more relevant question is whether we remain in a stage characterized by skepticism, or whether we have already entered a stage defined by fear of missing out.

No investor can answer that question with certainty. However, history suggests that every major technology infrastructure cycle follows a familiar pattern. Skepticism gives way to optimism, optimism attracts capital, and capital eventually creates additional supply. The transition from euphoria to normalization often occurs much faster than expected. While AI may ultimately transform large parts of the global economy, the investment cycle surrounding it is unlikely to be exempt from the laws of capital cycles.

What Meta's Compute Strategy May Be Signaling?

One recent development is worth highlighting. Meta's signal that it may lease spare AI compute capacity to outside users triggered a sharp market reaction and contributed to a pullback in the Philadelphia Semiconductor Index. The announcement itself mattered less than the signal it sent.

Historically, Meta was viewed as one of the largest consumers of AI computing power. If a major demand-side participant is beginning to behave like a supplier, investors naturally start reassessing long-term assumptions regarding compute demand growth.

At the same time, leading technology companies—including Google, Amazon, Microsoft, OpenAI, and Anthropic—are increasingly investing in proprietary AI inference chips. Should future compute demand become concentrated among a small number of frontier-model developers while supply continues to diversify through custom ASIC solutions, the distribution of profits across the AI value chain could change meaningfully over time.

Investors should pay close attention to these developments, as future industry economics may differ substantially from those implied by current market expectations.

China Equities: Broadening Beyond AI

Recent market action suggests that capital concentration is beginning to ease. As July got underway, many of the first-half leaders—communications equipment, electronics, and electronic materials—sold off meaningfully. Meanwhile, previously lagging sectors such as healthcare, agriculture, autos, and home appliances started to catch up. Market leadership appears to be broadening beyond a single AI-driven theme.

We continue to believe that the bull market that began in September 2024 remains intact.

However, the growing influence of quantitative strategies, passive investment vehicles, and ETF flows has amplified short-term market volatility. Sector rotations have become more aggressive, and portfolio fluctuations have become more pronounced than in previous cycles. In the short run, markets behave like a voting machine. Over the long run, though, they are still a weighing machine—and earnings growth is what ultimately drives value.

Recent regulatory measures aimed at reducing excessive capital concentration have also begun to support a more balanced market structure and healthier capital allocation.

Looking ahead to the third quarter, we expect market leadership to continue broadening. What began as a technology-led bull market may increasingly evolve into a more comprehensive market advance spanning multiple sectors.

Investment Implications

Our overall stance remains constructive.

The strength of the AI investment cycle is difficult to dispute. The more relevant question for investors is whether current valuations adequately compensate for the risks associated with future normalization. Distinguishing between industry growth and valuation expectations remains critical at this stage of the cycle.

Accordingly, we continue to favor a balanced approach:

· Maintain exposure to AI beneficiaries with durable competitive advantages and strong industry positioning.

· Remain disciplined on valuation, particularly in segments vulnerable to future capacity expansion.

· Seek opportunities beyond the most crowded trades as market leadership broadens.

· Focus on long-term value creation rather than short-term price movements.

Ultimately, investors would do well to spend less time watching daily P&L swings and more time assessing the quality and valuation of what they own. If a business remains fundamentally attractive and reasonably priced, short-term volatility is not the same as long-term risk. The market may vote in the short run, but over time it still weighs. That principle has held through every major cycle we've lived through—and we believe it will hold through this one, too. 

Wu Weizhi

4 July2026

本期《偉志思考》簡體中文版鏈接：

伟志思考：究竟是超级景气周期还是史诗级大泡沫