The storage chip industry bids farewell to the "boom-bust" curse! HBM is in short supply and price hikes have become the "new normal."

In the past few decades, the storage chip industry has continued to show a cyclical pattern of "boom-crash-boom", but now, the AI spending frenzy is pushing the storage chip industry into a new rhythm. Industry executives say that AI has structurally broken the old cycle, and there are no signs of prices decreasing. Hewlett Packard Enterprise Co. CEO Antonio Neri said, "We will continue to raise prices, as the entire industry will continue to raise prices. Currently, there is more demand than supply." A Seagate executive said on Tuesday that the rising prices of storage chips are likely to become the "new normal" in the coming years. South Korean storage chip giant SK Hynix said that the entire storage chip industry is undergoing structural changes. A spokesperson for the company stated, "Company customers, including large-scale data center operators, are increasingly inclined to sign long-term contracts, rather than the more common one-year agreements in the past." Micron Technology, Inc. also stated that customers are now very willing to sign long-term supply agreements to secure memory supply for the coming years. Broadcom Inc. CEO Tan Hock Eng said in the company's earnings call last week that he has secured supply until 2028. The current AI workloads require a completely different architecture from the past, with higher storage requirements, which far exceeds the industry's previous design foundations. Meta Platforms announced the launch of its own AI chip on Wednesday and expressed concerns about the supply of high-bandwidth memory (HBM) it requires. Meta's VP of Engineering, Song Yijun, said, "We are very concerned about the HBM supply. But we believe we have secured the necessary supply for the planned infrastructure." With large-scale cloud computing companies squeezing consumer-level supplies of HBM, and new capacity not expected to alleviate the situation until at least 2027, AI infrastructure development may have pushed the storage industry into a new era. Historical storage chip shortage By 2025, as global AI infrastructure development enters a boom period, the storage industry is experiencing a "super cycle." The key logic is that on one hand, AI servers require far more storage capacity and bandwidth than traditional servers. On the other hand, industry capacity is shifting towards high-end storage products (such as HBM), squeezing the capacity of traditional storage products and leading to an industry-wide price surge. Currently, AI infrastructure development is still accelerating, and large technology companies are expected to spend a staggering $650 billion in 2026, an increase of about 80% from last year's record level. According to industry research, data centers accounted for about 50% of global DRAM consumption in 2025, up from 32% five years ago. This proportion is expected to continue to rise. By 2030, AI servers are expected to account for over 60% of global storage demand. AI demand is causing a historic storage chip shortage. Meeting the exponentially growing demand for chips will be expensive, and may even be impossible to achieve. Although companies like Micron Technology, Inc., SK Hynix, and Samsung Electronics are expanding capacity through new construction or facility upgrades, such projects often take years and require billions of dollars in investment to yield significant output. HBM also presents additional challenges - it is extremely difficult in large-scale manufacturing. HBM is created by stacking multiple storage chips (each thinner than a hair) with micrometer precision. Any defect could ruin the entire stack, making production slower and yields lower than with traditional DRAM. Some versions of HBM also integrate small logic chips for managing and routing data, adding further complexity and consuming a significant amount of manufacturing capacity. Market research firm IDC bluntly stated that the storage chip shortage is becoming an "unprecedented crisis" as pressure from the AI frenzy mounts on supply. Executives including those from Apple Inc., Alphabet Inc. Class C, and Tesla, Inc. are discussing the impact of the storage chip shortage on profitability and even the timeline for AI advancements. Google DeepMind's Demis Hassabis called it an industry "bottleneck". During Tesla, Inc.'s earnings call in late January, CEO Elon Musk even proposed the idea of producing their own storage chips. Is the "storage super cycle" still on? Previous performance guidance from Micron Technology, Inc., as well as assessments from institutions such as Nomura and Citigroup, suggest that the current super cycle in the storage chip industry is expected to continue until 2026 or 2027. However, the recent actions of the "Wall Street Bear" Citron Capital targeting SanDisk have "poured cold water" on the current storage chip industry frenzy. Citron Capital's targeting of SanDisk is a reassessment of the prevailing logic of the "storage super cycle" in the market. This short-selling firm bluntly points out that the market's pricing logic for SanDisk is fundamentally flawed, and that the current tight supply of storage chips is just a "mirage", with the top of the cycle looming. Citron provided three main reasons. One of them is the curse of the storage cycle. Citron pointed out that the traditional storage chip industry has clear cyclical properties, with peaks in profitability seen in 2008, 2012, and 2018, and the current industry situation replicating history. Citron even explicitly stated that there is now twice the capacity on the market compared to the peak in 2018, and once released, it will completely reverse the supply-demand balance. Citron's bearish report has sparked doubts about the "storage super cycle" and caused short-term volatility in related stocks. Some analysts believe that starting from 2024, this round of storage cycle driven by AI is fundamentally different from past cycle-driving factors, essentially being the "AI storage chip cycle". Under the grand narrative of AI, companies with HBM as a high-end product moat have the confidence to transition from cyclical stocks to growth stocks. Standard storage products, still mainly used in smartphones and PCs, have not escaped strong cyclical tendencies, as production capacity at major factories contracts. If by 2027, overseas storage giants shift to expansion, there is still a risk of a cycle reversal in the industry, but this turning point has been significantly delayed by the explosion of AI demand.