"How long will the storage industry face shortages?" Unable to enter the exhibition hall, can't get into the venue. What trends did CFMS reveal this year?

Compared to the crowded scene in 2025, the Global Storage Industry Benchmark Summit CFMS｜MemoryS 2026 venue is quite different.

Around noon, the elevators leading to the second-floor main venue and exhibition hall were stopped, and access to the stairs was restricted; on-site security only allowed exhibitors and media to enter; in the main hall, audiences crowded at the entrance, and the temperature inside had risen a few degrees. The outdoor and indoor venues on the first floor were filled with people, and attendees were both afraid of missing any wealth secrets and worried about catching the last leg of this “storage super cycle.”

As the host, the Flash Market’s general manager, Tai Wei, reviews the market every year: In 2023, the industry was concerned about how much longer storage prices would drop, while in 2024, the focus shifted to how much longer prices would rise. “This year, even hotel staff are asking me how long the storage industry will continue to face shortages.”

According to estimates from the Flash Market organization, the global market size for semiconductor storage will exceed $600 billion in 2026, several times more than at any other time in the past. The focus of the storage industry has shifted from price comparisons to securing inventory, “some products are immediately plugged into customers’ servers right after they come off the production line, which was unimaginable in the past.”

“Our sales are busier than ever, and we have to check every day who has more urgent needs,” stated an exhibitor, indicating they might only meet 30-40% of customer demands and would directly refuse orders if demand was too high.

Various signals seem to point to the fact that this round of storage market growth is different from previous ones. However, everyone is also questioning another issue: how long will this level of prosperity last?

From a supply perspective, although storage manufacturers have begun to increase capital expenditures to expand production capacity, after several rounds of “mass loss cycles,” manufacturers no longer adhere to scale supremacy but rather “disciplined production increases,” prioritizing high-tech, high-value, and high-barrier products. This has also led to reduced supply elasticity, and future fluctuations are expected to narrow, potentially extending the industry prosperity cycle.

From a time dimension, the expansion cycle for storage capacity often lasts 18 to 24 months, with new capacity expected to be released no earlier than 2027, which is not a timely solution. Tai Wei pointed out, “In 2026, there will not be a single mainstream AI product that can achieve a complete balance of supply and demand.” It is predicted that storage product prices have already experienced three consecutive quarters of substantial increases, and it is expected that from the third quarter of 2026, the rate of increase will gradually converge, with some specific product prices showing differentiation.

Executives from the leading storage controller company, Phison, insist that 2026 will not be considered a dark moment; the supply-demand gap will widen in 2027. This round of price increases and shortages is not a simple cyclical fluctuation but rather a structural transformation driven by AI. After all, the massive amounts of data generated by AI training and inference create unprecedented demand for storage.

In this context, traditional storage giants like mobile phones will be the most affected. Major institutions have lowered their shipment forecasts for this year. It is predicted that due to rapidly rising costs, mobile market sales will decline by about 10%, with some phone models experiencing price drops of up to 30%.

In the face of cost pain for mobile manufacturers, many terminal and storage companies are also starting to pursue a “cost-performance” route, offering 4GB RAM equivalent to an 8GB experience, and increasing the procurement of domestic storage chips. Additionally, as AI focus shifts from training to inference, domestic manufacturers are exploring how to optimize systems through algorithms, design, and packaging to achieve the best results with the lowest power consumption and minimal latency on the edge AI.

International storage manufacturers are focusing more on cloud-side solutions, launching more advanced high-performance storage options: Samsung Semiconductor is accelerating the generational transition of PCIe interfaces and will introduce PCIe Gen6 solid-state drives this year, even advancing the technical reserves and product research for PCIe 7.0 and future PCIe 8.0; Kioxia is expanding solid-state drives by inserting a “dedicated KV Cache storage layer” between DRAM and SSDs to address HBM scalability and cost constraints.

As the super buyers in this storage cycle, cloud service providers have begun to stockpile storage “cost-performance” solutions.

Google recently announced the compression algorithm TurboQuant, which reportedly can reduce the key-value cache memory usage during large language model operations to about 1/6 of the original, improving performance by eight times. Industry insiders expect that Google’s new technology may ease the memory supply shortage, potentially lowering memory prices. Following this news, global storage stocks saw a significant decline this week.

“This is similar to the drop in Nvidia’s stock price after the emergence of DeepSeek. In the long term, improving efficiency is beneficial for storage, and AI demand for storage will not see major changes long-term,” stated a leader from a domestic storage company to reporters.