AI Memory Crunch: The Supercycle Has Only Just Begun

Deep News13:42

As global investors celebrate NVIDIA's earnings, a clear industrial signal has surfaced: the bottleneck in AI infrastructure has fully shifted to the memory segment. NVIDIA's management has explicitly stated that memory price increases have exceeded previous expectations, with supply constraints expected to persist through the end of fiscal 2028. A forecast from TrendForce paints an even clearer picture: by 2027, DRAM and NAND Flash will account for 68% of global cloud service providers' total capital expenditures, up from 47% in 2026. This means for every 100 million yuan a cloud provider invests in AI infrastructure, nearly 70% will flow into the memory supply chain.

With memory shortages now a key factor constraining the gross margins of cloud giants, storage chips have officially transitioned from being mere "supporting infrastructure" for computing power to becoming "strategic currency" for AI. As the starting point of this chip supercycle, the global storage sector has experienced volatile swings since the second half of the year, with clear battles between bulls and bears. As it stands now, this supercycle is far from over.

Natural Mismatch: Demand Shifts Fast, Supply Moves Slow

According to SEMI's World Fab Forecast and the publicly available construction timelines of Samsung, SK Hynix, and Micron, a new 12-inch memory fab typically takes 24 to 36 months from groundbreaking to stable mass production. Even for capacity expansions or process migrations within existing fabs, the journey from equipment installation to yield ramp-up takes 12 to 18 months. Capacity additions are measured in years, while demand shifts occur on a quarterly basis through enterprise contracts. This creates an inherent time lag. The full chain—from price signals to channel inventory to manufacturer capacity decisions to effective supply changes—typically takes over two years to play out. For memory manufacturers, when demand far outstrips supply, one critical element emerges: pricing power.

Clearly, the demand scale driven by the current AI computing infrastructure buildout far exceeds the industry's traditional expectations. Since the second half of 2025, memory contract prices have surged at an astonishing rate. TrendForce data shows server DRAM contract prices rose 64% cumulatively in the second half of 2025, with an additional ~270% increase expected in 2026. Enterprise SSD (NAND Flash) prices climbed about 35% in H2 2025, with an expected cumulative rise of 235% in 2026. Even HBM contract prices could still rise 70% to 140% in 2027.

Transformation: From Commodity Cyclical to High-Barrier Growth Stock

For decades, memory was viewed as a cyclical sector following classic commodity supply-demand logic. In traditional cycles, pulse-like demand surges would trigger blind capacity expansion, leading to oversupply when demand faded, prices crashed, and the industry entered prolonged inventory correction and capacity consolidation phases. This historical pattern is now being disrupted. First, AI model training, inference, and multimodal data processing all involve frequent data transfers between computing units and storage units, making memory an increasingly core component of computing performance. More importantly, AI demand is expanding from training to inference, shifting storage needs from "one-time investments" to "long-term, sustained computing infrastructure spending." This fundamentally distinguishes the current supercycle from past cyclical booms.

Second, supply-side capacity is heavily tilted toward high-end HBM, squeezing conventional DRAM production. Combined with shortages in core equipment and materials, plus the marginal impact of technology iterations on effective supply, multiple supply-side constraints mean the tightness will be difficult to resolve in the near term. Furthermore, business models are evolving: long-term agreements are delivering exceptional earnings visibility. Samsung, SK Hynix, and Micron have already allocated their full 2027 capacity. Micron holds 16 long-term strategic partnership orders, 14 of which guarantee over 100 billion yuan in baseline revenue. Given this long-term structural shortage and technological irreplaceability, rather than questioning how long memory price hikes can last, it's more accurate to recognize that storage has transformed into a high-barrier growth sector.

Opportunities for Domestic Chinese Memory Players

The reshaping of the global memory industry coincides with a historic window for Chinese memory companies as they transition from technological breakthroughs to large-scale production. Based on Q4 2025 DRAM sales, ChangXin Memory's global market share has risen to 7.67%, ranking fourth worldwide. According to Counterpoint Research's NAND memory market tracker, YMTC's global market share jumped to 13% in Q1 2026. In the general-purpose memory market, domestic Chinese players have entered the global top tier, but in the high-end HBM segment, a significant gap remains that will take time to close.

The overseas giants' capacity shift toward HBM is unlikely to reverse in the near term. ChangXin's DDR4, DDR5, and LPDDR5X products are already in full-series mass production, while LPDDR6 has completed sampling with core customers and is planned for mass production introduction in H2 2026—perfectly positioned to absorb the mid-range general DRAM market vacated by overseas players. Meanwhile, AI-computing storage products have been added to the Xinchuang (domestic substitution) procurement catalog, with government and enterprise server storage substitution potential reaching the 100-billion-yuan level. With the third phase of the National Big Fund and various industrial policies providing support, the domestic storage supply chain, anchored by ChangXin and YMTC, is accelerating its breakthrough. ChangXin has secured multi-billion-yuan long-term orders from ByteDance, Tencent, and Alibaba Cloud, while YMTC's enterprise SSDs are rapidly entering the supply chains of domestic AI computing centers.

Three Main Investment Themes

The spillover effects of the memory supercycle are spreading across the entire supply chain. According to aggregated broker views, three core themes stand out: first, memory manufacturers and module makers with high-end storage interface design capabilities or successful entry into top cloud providers' supply chains; second, high-end memory-specific equipment companies directly benefiting from order expansion; and third, advanced packaging and testing and specialty materials suppliers. UBS notes that while Chinese memory makers lag in DRAM technology nodes, their international market share will continue to rise significantly, with domestic substitution in the data center space being particularly evident. Domestic semiconductor equipment, especially front-end and back-end testing equipment, will benefit first from memory capacity expansion. Additionally, the favorable supply-demand dynamics in mature process nodes, high utilization rates, and rising prices will support a notable recovery in net margins for domestic wafer fabs.

*Related ETF: The HuaBao Shanghai Sci-Tech Innovation Board Chip ETF (589190) passively tracks the Shanghai Sci-Tech Innovation Board Chip Index, providing full-chain exposure to the chip industry. Memory, semiconductor materials, and equipment account for 49.95% and 30.57% respectively (as of 2026.8.26), effectively capturing the industrial development trends driven by the continuous upgrade of AI computing infrastructure. With high hard-tech content and strong offensive characteristics, it offers 20CM high elasticity. Off-market investors may also consider the feeder fund 021225.

Data source: Shanghai and Shenzhen stock exchanges, etc. Note: The memory chip exposure corresponds to the Memory Chip Index (980138.CNI), with a weight of 49.95% as of August 26th. Institutional opinion source: UBS Securities 20260827, "Chinese Memory Manufacturers' Global Shipment Share Will Continue to Rise."

ETF fee disclosures: When investors subscribe for or redeem fund shares, the subscription and redemption agency may charge a commission of up to 0.5%, which includes fees charged by stock exchanges, registration agencies, and other relevant entities. Feeder fund fee disclosures: For the HuaBao Shanghai Sci-Tech Innovation Board Chip ETF Feeder Fund A, the subscription fee (front-end) is 1000 yuan per transaction for amounts of 2 million yuan (inclusive) or more, 0.2% for amounts between 1 million (inclusive) and 2 million yuan, and 0.5% for amounts below 1 million yuan. The redemption fee is 1.5% for holding periods of less than 7 days and 0% for holding periods of 7 days (inclusive) or more. The Feeder Fund C class does not charge a subscription fee, has a redemption fee of 1.5% for holdings less than 7 days and 0% for 7 days or more, and a sales service fee of 0.2%.

Risk disclosures: The Sci-Tech Innovation Board Chip ETF (589190) and its feeder fund passively track the Shanghai Sci-Tech Innovation Board Chip Index. The index base date is 2019.12.31 and its publication date is 2022.6.13. The return of the Shanghai Sci-Tech Innovation Board Chip Index over the past 5 complete years was 6.87% (2021), -33.69% (2022), 7.26% (2023), 34.52% (2024), and 61.33% (2025). The volatility over these periods was 34.32% (2021), 36.60% (2022), 28.64% (2023), 44.67% (2024), and 34.34% (2025). Index constituents are adjusted periodically according to the index compilation rules. Backtested historical performance does not indicate future index performance. This product is issued and managed by HuaBao Fund. Distribution institutions do not bear investment, redemption, or risk management responsibilities. Investors should carefully read fund legal documents including the Fund Contract, Prospectus, and Fund Product Information Summary to understand the fund's risk-return characteristics and select products suited to their own risk tolerance. The fund's risk rating assessed by the fund manager is R4 (medium-high risk), suitable for investors with suitability rating C4 or above. The performance of other funds managed by the fund manager does not constitute a guarantee of this fund's performance. Past fund performance does not indicate future returns. Funds involve risks; investors should invest cautiously!

Sales institutions (including the fund manager's direct sales and other sales channels) conduct risk assessments of this fund according to relevant laws and regulations. Investors should promptly pay attention to the suitability opinions issued by the fund manager. Suitability opinions from various sales channels may not be consistent, and the fund product risk level assessments issued by fund sales institutions must not be lower than those issued by the fund manager. The fund contract's description of risk-return characteristics may differ from the fund risk rating due to varying considerations. Investors should understand the fund's risk-return profile and carefully choose fund products based on their own investment objectives, horizons, experience, and risk tolerance, assuming their own risks. The registration of this fund with the China Securities Regulatory Commission does not imply its substantive judgment or guarantee of the fund's investment value, market prospects, or returns. Funds involve risks; invest cautiously. A MACD golden cross signal has formed, and these stocks are performing well!

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