Sinolink Securities: AI-Driven New Storage Cycle – Historic Opportunity Amid Demand Surge and Supply Bottlenecks

Stock News11-26

Sinolink Securities released a research report stating that historical cycles show the storage market exhibits strong cyclicality, with each major cycle being driven by emerging technologies. The industry now stands at the starting point of a new storage supercycle. Investors are advised to closely monitor inventory levels, pricing trends, and the demand boost for memory chips driven by AI computing power. Key insights from the report include:

Semiconductor cycles can be categorized into long (8–10 years, demand-driven), medium (4–6 years, capacity-driven), and short (3–5 quarters, inventory-driven) cycles, with these three dimensions interwoven. Memory is the second-largest segment in semiconductors after logic, historically aligning with broader semiconductor cycles but displaying higher volatility. Major storage cycles (e.g., 2008, 2016) were triggered by technological innovations that upgraded products, expanding total demand, penetration rates, and memory value—ultimately elevating market scale. With AI fueling demand, a new supercycle is emerging.

**Will AI Drive a Storage Supercycle?** 1) **Exploding Data Storage Needs**: Large language models (LLMs) now employ chain-of-thought prompting, breaking complex problems into actionable steps and mimicking human reasoning. This enhances model performance but increases token consumption during inference. By 2025, most leading models will integrate this mechanism. Meanwhile, Seagate data shows storage demands leap from KB to TB/EB as content shifts from text to audio/video. Multimodal model adoption will further accelerate storage needs.

2) **Cost Reduction Spurs Demand**: Since ChatGPT-3’s launch, LLM inference costs have dropped exponentially, with tokens generated per dollar rising steadily. Lower costs could trigger an application boom, indirectly boosting storage demand.

3) **AI Redefines Storage**: KV Cache, a Transformer-based optimization for inference efficiency, scales linearly with token count and text length. As KV Cache becomes critical for model performance, storage requirements will rise.

**Supply Constraints Amid Demand Surge** DRAM and NAND Flash manufacturers are pivoting from pure capacity expansion to higher-value innovations like advanced nodes, 3D stacking, hybrid bonding, and HBM. TrendForce projects DRAM capex to hit $53.7B in 2025 (+14% YoY to $61.3B in 2026), while NAND capex reaches $21.1B in 2025 (+5% YoY to $22.2B in 2026). Cleanroom space is nearing limits, with only Samsung and SK Hynix retaining marginal expansion room. Even with capex hikes, 2026 output growth remains constrained. In NAND, Kioxia/SanDisk leads expansion, while others prioritize HBM and DRAM.

**Risks**: Weaker-than-expected demand, memory price declines, or price hikes crowding out other sectors.

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Comments

We need your insight to fill this gap
Leave a comment