To address bandwidth bottlenecks and cost pressures, the storage industry needs to innovate through six major approaches including advanced packaging. At the 2026 Global Memory Industry Innovation Summit (GMIF2026) held in Shenzhen on September 23, Yan Zhitian, Executive Director of Morgan Stanley, said during his speech that global cloud service providers' (also known as "cloud") capital expenditure is expected to reach $1.2 trillion in 2027, with more than half allocated to storage. He predicted that some cloud service providers' capital expenditure next year will exceed their EBITDA, and because storage is expensive and bandwidth is insufficient, innovation needs to be accelerated in areas such as advanced packaging.

Morgan Stanley research reports show that global cloud capital expenditure this year may exceed $900 billion, with storage (excluding HBM (High Bandwidth Memory)) accounting for 48%; next year global cloud capital expenditure may reach $1.2 trillion, with storage's share increasing to 53%; the year after, as demand for storage and GPUs grows, global cloud capital expenditure will reach $1.6 trillion.

Under the AI wave, strong demand has driven storage prices sharply higher. Under these circumstances, the ratio of 2026 capital expenditure to EBITDA for the four major cloud service providers — Google, Amazon, Microsoft, and Meta — has already exceeded 70%, and in 2027 some companies' capital expenditure will exceed EBITDA. This has triggered discussions in the capital markets, and some related companies have recently faced valuation pressure.

"A very hot topic right now is how much HBM demand there will be next year," Yan Zhitian said. Preliminary estimates suggest that next year's HBM consumption will be approximately $44 billion, a significant increase from last year and this year, but supply is still very tight at present. DRAM (Dynamic Random Access Memory) and NAND (flash memory) also faced tight supply in the second half of this year, with the DRAM supply-demand gap possibly around 15%; the NAND supply-demand gap may be close to 10%, with enterprise-grade memory facing a larger supply-demand gap while consumer-grade memory's supply-demand gap is relatively narrowing.

Due to the supply shortage, storage is currently very expensive, accounting for approximately 73% of CPU server costs, 41% of laptop costs, and 39% of smartphone costs. On one hand, storage is expensive; on the other hand, bandwidth is limited. Yan Zhitian said that comparing Nvidia's AI computing chip B200 with high-bandwidth memory HBM3 and HBM4, their bandwidth differs by 20,000 times, so storage urgently needs innovation.

Yan Zhitian believes that there are six major approaches to storage industry innovation: process evolution, storage design improvement, advanced packaging, adding companion chips, compute-in-memory, and material improvement. By 2030, the market size for storage innovation (excluding HBM) will reach $25 billion, with the advanced packaging and peripheral companion chip markets growing relatively quickly.

At the summit, storage companies also shared their innovation ideas. Samsung Electronics Vice President and Chief Technology Officer of the Memory Business, Yin Kaiwen, said that as agentic AI scales up, Samsung is optimizing memory architecture by keeping the operating system and KV cache in DRAM and storing large-capacity models in Z-NAND, enabling trillion-parameter large models to be deployed on edge AI. SanDisk Senior Product Marketing Director Zhang Dan also believes that local AI will be the trend because it deploys faster, is more private, and more controllable, so different storage products are needed to adapt to different application scenarios.