Three and a half years into the Generative AI boom, the global market has reached a critical inflection point where accelerating optimism clashes with accumulating skepticism. On June 9, the South Korean KOSPI index staged a volatile rebound with an intraday gain of nearly 5%, triggering a circuit breaker after falling over 8% the previous day to drop below 8000 points. This volatility underscores South Korea's role as a primary amplifier for global AI sentiment, where stocks like NVIDIA and HBM have surged alongside expanded server production benefiting SK Hynix, while storage price hikes have rewritten valuation logic for Samsung and Micron. The market's repeated activation of trading cooling-off mechanisms reflects a deepening divergence in global capital regarding the sustainability of AI infrastructure expansion.

The core contradiction lies in the dual reality of the sector: AI remains the most certain investment theme, yet skepticism is mounting as costs escalate. From chips to large-scale models, core assets are re-integrated into an 'AI infrastructure' valuation framework where current capital expenditures and supply chain price hikes are justified as upfront investments for future growth.

However, Data compiled by Woofun AI shows that tech giants are increasingly cash-flow negative, with the big four collectively burning $20 billion a day while reporting record profits. This divergence suggests that while demand for computing power drives valuations, the financial sustainability of this trajectory is becoming a focal point of market anxiety.

Historical analysis reveals that serious discussions about an AI bubble have already occurred in three distinct waves, each triggered by specific paradigm shifts and resolved by new market narratives. The first wave emerged in June 2024 when Sequoia Capital questioned the $600 billion annual revenue required to support infrastructure costs under the Pre-training Scaling Law. This doubt was quickly quelled by the September 2024 release of OpenAI's o1, which introduced a reasoning-time computing paradigm that shifted focus from model size to thinking duration. The second wave began on January 27, 2025, when NVIDIA's market capitalization evaporated by $593 billion in a single day following the release of DeepSeek R1, which achieved cutting-edge inference capabilities with less than $6 million in training costs. The market stabilized one month later when NVIDIA reported Blackwell quarterly revenue of $11 billion, proving that the new inference paradigm would drive even higher computing power requirements.

The third wave of skepticism peaked between October and November 2025, marked by a Goldman Sachs report identifying five bubble signs including peaking CapEx and slowing enterprise profit growth. Bank of America fund managers issued their first 'overinvestment' judgment in 20 years, while investigations by Wired and The Atlantic found that 95% of enterprise AI investments had not yielded tangible returns. Despite these warnings, major U.S. tech giants responded to analysts by stating they would rather overinvest than lose the future. Woofun AI notes that this consensus was reinforced by the advent of Agentic AI in late 2025, which transformed AI from dialogue tools to autonomous digital employees, theoretically increasing token consumption by tens of times and opening an order-of-magnitude growth space for computing demand.

The current fourth wave of divergence is characterized by a dangerous consensus where doubts are expected to be quickly refuted, leading to heightened risk appetite among investors. In the 2026 Q1 earnings season, a stark rift appeared between reported profits and actual cash flows; Amazon's free cash flow plummeted by 95% year-on-year, and Goldman Sachs estimated that 40% of the expected 2026 S&P 500 earnings growth stems from AI-related capital investment transmission. Morgan Stanley projects that by 2026, the CapEx-to-revenue ratio for large enterprises will reach 34%, surpassing the 32% historical peak of the 2000 Internet bubble, with total AI infrastructure expenditure for the top five giants reaching $20 trillion between 2026 and 2028.

Additionally, these companies hold nearly $1 trillion in off-balance-sheet lease commitments for data centers, further obscuring the true financial burden.

A significant fault line has emerged regarding the efficacy of the 'tokenmaxing' movement, where systemic redundancy within Agent architectures may be inflating usage metrics without corresponding productivity gains. Reports indicate that Uber burned through its entire annual AI coding budget in the first four months of 2026, as engineers utilized tools like Claude Code for parallel labor tasks that generate massive token usage but lack quantifiable output. Woofun AI analysis suggests that if usage volume metrics are bloated by architectural idling, the trillion-dollar valuations built upon them face severe reliability risks. This uncertainty is compounded by the fact that no institution has yet disaggregated the ratio of effective computation to architectural idling.

Capital markets continue to frenzy with Anthropic nearing a $965 billion valuation and OpenAI filing for an IPO at $852 billion, indicating the market is paying full price for an unrealized future. While Ray Dalio argues that bubbles burst when investors attempt to convert paper wealth into cash, Peter Thiel has already exited his $100 million NVIDIA position, accounting for 40% of his fund, and cut 76% of his Tesla holdings. Similarly, Berkshire Hathaway's Q1 2026 report shows Buffett's cash position ballooning to $397.4 billion, representing 59% of total assets. These moves highlight a growing contradiction between the long-term positive trend of technological evolution and short-term strategies focused on profit-taking and deleveraging.

As U.S. interest rate expectations rise and questions about overheated AI spending intensify, the market enters a highly challenging range where faith coexists with extreme volatility. The South Korean market's sensitivity serves as a barometer for this global tension, where sharp drops stem directly from loosening AI faith. While experienced investors may navigate this storm to find value, the need for vigilance is paramount as the fourth wave of the Generative AI bubble theory gains traction. The ability to distinguish between a directional judgment on technology and a comprehensive test of investment rhythm, position, and exit windows will define the next phase of the industry.