记忆体晶片行业有产能过剩的历史惯性，类似于水泥行业的周期规律。当价格高涨时，企业倾向于大规模扩建产能，但当景气下滑时，这些多余产能便成为沉重负担。三星被视为闪存产能过剩的主要推手，而SK海力士近期透过美国存托凭证募集了280亿美元用于新产能建设，令市场再度想起过去的记忆体晶片周期。此外，美光选择九月作为财年结束日，使其财务报表在库存和现金方面呈现较为有利的状态，可能掩盖了公司真实的财务风险。
展望未来，如果人工智慧需求的热潮能够持续多年，高效能记忆体制造商将获得超出常规周期的丰厚回报，新增产能建设也将合理。但若AI超大规模客户的营收成长不足以支撑庞大的资本支出，繁荣可能迅速转为萧条。作者提醒投资者回顾2000年网路泡沫的教训——当年网路入口网站股价见顶后，热潮转向网路基础设施建设商，而当这些股票也停止创新高时，泡沫便彻底破灭。
The semiconductor industry is inherently cyclical, and recent events vividly illustrate this pattern. Despite reporting extraordinary profit growth—Samsung's quarterly profits surged 19-fold and Micron's 15-fold—their share prices fell sharply. This counterintuitive behavior is characteristic of cyclical stocks: price-to-earnings ratios appear deceptively low at earnings peaks (signaling a time to sell) and infinitely high during downturns when earnings turn negative (paradoxically signaling a buying opportunity), as demonstrated by Micron's share price history over the past 15 years.
The memory chip sector has a well-documented tendency toward overbuilding capacity during boom times, much like the cement industry. Samsung has historically been associated with excess flash memory capacity expansion, and SK Hynix recently raised $28 billion through an ADR placement specifically to fund new capacity construction—echoing previous cycle patterns. Additionally, Micron's choice of a September fiscal year-end raises questions, as it tends to present the company's financial position at its most favorable point, with lower inventory and higher cash than typical levels, potentially masking the true extent of financial risk and leverage in such a cyclical business. (Key numbers: 280)
Looking ahead, the critical question is whether the AI-driven demand boom will sustain long enough to justify the massive capacity investments now underway. If AI hyperscalers generate sufficient revenue growth to absorb the enormous capital expenditure on chips, the cycle may be extended. However, if their revenue growth proves merely strong rather than strong enough to cover the bills, the boom could rapidly turn to bust. The author draws a cautionary parallel to the dot-com bubble of 2000, when the collapse of internet portal stocks was followed by the fall of network infrastructure builders like Cisco, suggesting that when the current AI hardware stocks stop making new highs, the cycle may be definitively over.