Post

Why are storage prices still rising, but storage stocks have already fallen? Because the stock market looks further ahead. Everyone basically knows how high the profits are today; the biggest controversy now is whether there will be an oversupply in two to three years. This scenario has already played out once in the last round of new energy vehicles. In 2021, demand for new energy vehicles exploded, global lithium mine supply couldn't keep up, and battery-grade lithium carbonate prices rose from about ¥60,000/ton to nearly ¥600,000/ton, increasing nearly 10 times in two years. In the industry chain, whoever is the scarcest gets the profits concentrated there first. Lithium mining companies made huge profits, while battery manufacturers and car companies had to bear increasingly high raw material costs. In 2022, CATL's gross margin once dropped from nearly 28% to about 15%, simply because lithium prices rose too fast and battery price increases couldn't keep up. High profits quickly attracted a lot of capital. Mine expansions, rising capital expenditures, and increasing long-term purchase agreements. The market also began to worry whether lithium would still be so scarce when these new capacities come online in two to three years. Therefore, lithium mining stocks often start to fall before lithium prices actually peak. When lithium carbonate prices plummeted in 2023, many thought car manufacturers could finally turn all cost reductions into profits. But the auto industry immediately entered a price war. Tesla cut prices, BYD followed, and more brands competed for market share. Batteries did get cheaper, but the saved money did not all stay with the car manufacturers; a large part eventually turned into lower car prices. Raw material cost reductions and downstream profit improvements are separated by industry competition dynamics. Today's AI industry is somewhat like the new energy vehicle industry back then. Storage manufacturers correspond to lithium miners, and cloud providers correspond to car manufacturers. Over the past year, prices for HBM, DRAM, and enterprise SSDs have continuously risen, significantly improving profits for storage manufacturers like SK Hynix, Micron, and Samsung. Meanwhile, AWS, Azure, Google Cloud, and Oracle have been increasing purchases of GPUs, HBM, and servers, pushing infrastructure costs higher. At this stage, the scarcest segment takes the most profit first. But the market has already started looking ahead. In the next two to three years, if HBM, DRAM, and advanced packaging continue to expand production, how long can this round of excess profits last? This is also why storage companies' earnings are still good, but their stock prices have already begun to adjust. However, storage peaking does not necessarily mean cloud providers will be the biggest winners in the next round. Because on the large model side, the price war has actually already begun. OpenAI, Google, Anthropic, as well as Alibaba, DeepSeek, and Moonshadow, are continuously lowering model prices. Tokens are getting cheaper, inference costs keep dropping, and some models are even offered for free. If computing power supply becomes increasingly abundant in the future, cloud providers may continue to cut prices to compete for customers. By then, cost improvements from storage price drops may not fully translate into profits for AWS, Azure, or Google Cloud. Cheaper tokens, lower GPU rental prices, and larger free quotas may pass these benefits on to customers. So this round of storage stock adjustments can be understood through the lens of the new energy vehicle cycle: When upstream resources are scarcest, profits concentrate upstream first; high profits stimulate expansion, and stock prices start worrying about supply release in advance; when raw material prices really fall, how much profit downstream can keep depends on whether the industry has started a price war. AI is already showing these signs now.