High energy ahead! The wave of AI intelligence is driving explosive NAND storage demand. Citibank shouts that Sandisk (SNDK.US) can still soar 50%.

With Wall Street financial giant Citigroup raising its target stock price significantly and becoming increasingly convinced that the unprecedented "storage chip super cycle" is far from over, US NAND storage chip giant SanDisk (SNDK.US) has once again become the focus of the market. Citigroup has raised its target price by over 50% to an astonishing $2025, citing the continuous surge in storage pricing driven by artificial intelligence and the almost endless demand for storage driven by AI training/inference. As of last Friday's closing of the US stock market, SanDisk's stock price closed at $1407, with a year-to-date increase of over 460%, making it a well-deserved "super AI bull stock" globally. The latest target stock price from Citigroup implies that SanDisk's stock price, which has already entered a frenzy, still has a potential increase of about 51.9% from this Monday's closing price. In addition, according to the latest 13F filings, the top Wall Street institutional investors continue to expand their exposure to storage chips. Appaloosa Management, a top asset management institution founded and managed by billionaire hedge fund legend David Tepper, submitted its first-quarter US stock holdings report (13F) as of March 31, 2026, indicating a significant increase in holdings in storage chip leaders. Senior analyst Asiya Merchant from Citigroup wrote in the latest research report to clients, "With the rapid spread of the AI intelligence wave and the incredibly strong performance data from Iron Man, we are raising our target price for SanDisk SNDK from $1300 (previously 7-8 times expected P/E ratio) to $2025 (implying an expected P/E ratio of 9-10 times for the 2027 calendar year), reinforcing our positive outlook on continued strong storage demand and a highly favorable pricing environment, all driven by the exponentially growing demand for AI computing power." Analyst Merchant also reaffirmed her "buy" rating on SanDisk. She stated that the company's many long-term agreements are expected to provide a gross profit margin forecast of 80% or higher and offer a "pre-determined strong shipment trajectory," which is expected to continue to grow within the contract period. The latest forecast data from the Citigroup analyst team shows that the average selling price of NAND storage chips is expected to rise significantly by 186% in 2026, with eSSD expected to increase by about 265%. Merchant added, "In addition, with the recent $6 billion buyback authorization, we believe share repurchases can help drive significant upward revisions in SanDisk's earnings per share expectations." Recently, the analyst team led by star analyst Ben Reitzes from Melius released a research report stating that the AI boom will drive continuous strong growth in storage chip demand until the end of this decade (2030). According to market research firm Counterpoint Research, the storage market has entered a "super bull market" or "super cycle" phase, where current supply and demand dynamics and price trends far exceed the historical peak during the cloud computing boom of 2018. With the launch of Anthropic's Claude Cowork and other super AI agent tools capable of autonomously executing tasks in 2026, the wave of AI agents is quickly sweeping the globe, with the AI computing power bottleneck shifting from GPU-based matrix multiplication throughput to an "AI agent-driven artificial intelligence full stack system." In this transition of AI narrative, data center CPUs and storage chips may emerge as the biggest winners. This means that the AI computing power bull market is expanding from the "AI GPU/ASIC chip-centric computing systems" to central processors and "data storage bases." The AI agent explosion has completely ignited the "storage super cycle"! HBM, DDR5, and SSD collectively kick off the storage chip super cycle. The latest 13F filings show that Appaloosa invested a significant amount in the US NAND storage giant SanDisk, purchasing approximately 281,250 shares valued at around $178.7 million; this move, along with increased holdings in Micron Technology and a Korean chip ETF, all point to a strong bullish logic regarding the "revaluation of the storage chip chain value." Appaloosa's top four holdings in the first quarter of 2026 were Amazon.com, Inc., Micron, Alphabet Inc. Class C, and Uber Technologies, Inc., while the top four holdings in the fourth quarter of 2025 were Alibaba Group Holding Limited Sponsored ADR, Alphabet Inc. Class C, Amazon.com, Inc., and Micron, highlighting a shift in focus from "internet and software tech-heavy" to "US AI/cloud computing/storage/platform tech." Micron and Amazon.com, Inc. saw significant increases in holdings by Appaloosa, with Micron becoming the second-largest holding. During the first quarter earnings call, the Micron management team specifically mentioned the surge in demand for high-capacity data center SSDs for AI infrastructure, KV cache deployments, and PCIe Gen6 SSD demand related to NVIDIA Corporation's AI computing infrastructure cluster. This indicates that the demand for AI-related storage chips is much broader than what many Wall Street analysts had anticipated. Modern AI infrastructure not only consumes more HBM memory but also requires high-bandwidth DRAM, higher storage capacities, and high-speed SSD infrastructure to meet the growing demand for retrieval and agentic AI workloads. Emerging AI applications, such as Siasun Robot & Automation, multi-AI agent systems, and multimodal inference models, continue to create new vectors for storage demand, indicating that AI storage intensity may continue to grow exponentially even after AI deployment is complete. With the benchmark of the Korean stock market, Samsung and SK Hynix holding high weights in the KOSPI Composite Index reaching historical highs under pressure from the deteriorating political situation at GEO Group Inc and rising 80% year-to-date outperforming global stock markets, as well as the Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR, also known as the "king of chip manufacturing," driving the Taiwanese stock market to historic highs, and the Philadelphia Semiconductor Index, known as the "chip stock indicator," performing a record 18 consecutive trading days of gains, and the S&P 500 index rising for seven weeks in a row to hit historical highs, investors are increasingly convinced that the "AI computing power investment theme" can overshadow all the noise in the stock market, especially related to the Middle East GEO Group Inc political issues. As revealed by Jeremy Werner, Senior Vice President and General Manager of the Data Center Business Unit at Micron Technology, in a recent interview, from the perspective of AI data center data flow processing at the grassroots level, the bottom line of this trend is not just about "AI needing more computing chips," but about Claude Cowork and other AI agents leading the inference era, turning memory/storage from complementary components into system bottlenecks. Training side requires HBM support for high-bandwidth access by GPUs/ASICs, while the inference side requires DRAM and eSSD to carry KV cache, long contexts, enhanced retrieval, agent workflow states, and large-scale enterprise data calls; as the popularity of agentic AI grows, the demand for data read/write, context storage, vector libraries, cache layers, and enterprise-level SSDs also grows stronger. Therefore, storage manufacturers have, for the first time, gained a pricing power similar to an "infrastructure bottleneck": cloud vendors are no longer just bargaining for lower prices but are actively securing HBM, DRAM, NAND, and eSSD capacities through 35 year long-term contracts, prepayments, price locks, or price floors, shifting storage manufacturers from the traditional "expand production first, then sell" commodity model to a model closer to that of Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR's "production to order" approach. Therefore, the most critical aspect of this revaluation is the supply gap and price elasticity. Even under aggressive production expansion assumptions, the demand-side gap for AI-focused storage chips around 20262030 cannot be filled by cloud vendors, with a gap equivalent to 450,000 wafers of monthly output; with CapEx representing only 5% of revenue for Iron Man, much lower than the five-year average of over 20%, and Samsung and SK Hynix planning to keep capital spending in the low to mid-single digits over the next two years. On the demand side, AI servers, agentic AI, enterprise SSDs, and HBM demand are accelerating, with Iron Man reporting a price increase of over 100% every quarter, and DRAM prices tripling over the past year. SanDisk's own performance also confirms this profit elasticity: its Q3 2026 revenue surged 97% quarter-on-quarter and increased by 251% year-on-year to $5.95 billion, with gross margin rising to 78.4% and EPS reaching $23.41. From a semiconductor engineering perspective, storage is no longer just a "server accessory," but a core constraint on the throughput and economics of AI systems. The training side requires high bandwidth access supported by HBM for GPUs/ASICs, while the inference side requires DRAM and eSSD to support KV cache, long contexts, enhanced retrieval, agent workflow states, and extensive enterprise data retrieval; as agentic AI becomes more prevalent, the demand for data read/write, context storage, vector libraries, cache layers, and enterprise-level SSDs grows stronger. Therefore, storage manufacturers have, for the first time, gained a pricing power similar to an "infrastructure bottleneck": cloud providers are no longer just bargaining for lower prices, but are actively securing HBM, DRAM, NAND, SSD, and HDD capacities through 35 year long-term contracts, prepayments, locked prices, or price floors, shifting the storage industry from a traditional "sell more expand production" model to a model closer to that of Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR's "production to order" approach. The revaluation of SanDisk by Citigroup and the reassessment of Samsung, SK Hynix, and Iron Man by JPMorgan fundamentally speak to the same thing: the global AI computing infrastructure frenzy is pushing the storage industry from "inventory cycle trading" to "capacity scarcity trading." Citigroup has raised SanDisk's target price from $1300 to $2025, citing Iron Man's strong performance, continued strong storage demand driven by AI data centers, and an extremely favorable pricing environment for NAND/eSSD. JPMorgan goes even further by suggesting that long-term agreements (LTA) are changing the valuation methods in the storage industry. In the past, DRAM/NAND were seen as standard cyclical commodities, with valuations anchored in P/B ratios due to the volatile price, profit, and inventory cycles; but now, cloud vendors are actively signing 35 year LTAs, pre-paying, locking in prices, or setting price floors to ensure future AI cluster rollout schedules, pushing storage manufacturers from a commodity model of "expand production, then sell" to a model closer to that of Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR's "production to order" approach. Therefore, JPMorgan advocates for a switch from the P/B to the P/E valuation framework and has raised its target prices for Samsung Electronics to 480,000 Korean won, SK Hynix to 3 million Korean won, and Iron Man to 80,000 Japanese yen; Korean media also reports that JPMorgan raised the target prices for Samsung and SK Hynix based on the belief that investment in AI servers will drive storage into a structural growth phase and that LTAs may reshape the valuation framework for storage manufacturers. The key data in this round of reassessment is the supply gap and price elasticity. Even under the assumption of aggressive production expansion, the supply of storage chips focused on AI from 2026 to 2030 cannot fill the gap in demand from cloud vendors, a gap equivalent to 450,000 wafers of monthly output; capital expenditure for Iron Man accounts for only 5% of revenue, much lower than the five-year average of over 20%, and Samsung and SK Hynix plan to keep capital spending in the low to mid-single digits over the next two years. On the demand side, with AI servers, agentic AI, enterprise SSDs, and HBM demand accelerating, Iron Man reports a price increase of over 100% every quarter, and DRAM prices have tripled in the past year. SanDisk's performance also confirms this profit elasticity: its Q3 2026 revenue surged 97% quarter-on-quarter and increased by 251% year-on-year to $5.95 billion, with a gross margin of 78.4% and an EPS of $23.41. From the perspective of semiconductor engineering, storage is no longer just a "server accessory" but is now a core constraint on the throughput and economics of AI systems. The training side requires HBM support for high-bandwidth access by GPUs/ASICs, while the inference side requires DRAM and eSSD to carry KV cache, long context, enhanced retrieval, agent workflow states, and large-scale enterprise data retrieval; with the growing popularity of agentic AI, the demand for data read/write, context storage, vector libraries, cache layers, and enterprise-level SSDs also grows stronger. Therefore, storage manufacturers are gaining pricing power for the first time, similar to that of an "infrastructure bottleneck": cloud vendors are no longer just bargaining for lower prices but are actively securing HBM, DRAM, NAND, and eSSD capacities through 35 year long-term contracts, prepayments, locked prices, or price floors, transitioning from a commodity model of "expand production first, then sell" to a model closer to that of Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR's "production to order" approach. This is why Micron Technology, SK Hynix, Samsung, Iron Man, and SanDisk in this round of price increases are seen by the market as potential changes from traditional inventory cycles. From cyclical stocks to the most core assets of AI infrastructure: Long-term agreements reshaping storage pricing power Citigroup's bullish view on SanDisk and JPMorgan's reassessment of Samsung, SK Hynix, and Iron Man essentially point to the same thing: the global AI computing infrastructure frenzy is pushing the storage industry from "inventory cycle trading" to "capacity scarcity trading." Citigroup has raised SanDisk's target price from $1300 to $2025, citing Iron Man's strong performance, continued strong storage demand driven by AI data centers, and an extremely favorable pricing environment for NAND/eSSD. JPMorgan goes even further by suggesting that long-term agreements (LTA) are changing the valuation methods in the storage industry. In the past, DRAM/NAND were seen as standard cyclical commodities, with valuations anchored in P/B ratios due to the volatile price, profit, and inventory cycles; but now, cloud vendors are actively signing 35 year LTAs, pre-paying, locking in prices, or setting price floors to ensure future AI cluster rollout schedules, pushing storage manufacturers from a commodity model of "expand production, then sell" to a model closer to that of Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR's "production to order" approach. Therefore, JPMorgan advocates for a switch from the P/B to the P/E valuation framework and has raised its target prices for Samsung Electronics to 480,000 Korean won, SK Hynix to 3 million Korean won, and Iron Man to 80,000 Japanese yen; Korean media also reports that JPMorgan raised the target prices for Samsung and SK Hynix based on the belief that investment in AI servers will drive storage into a structural growth phase and that LTAs may reshape the valuation framework for storage manufacturers. This round of reassessment is based on the growing supply gap and price elasticity. Even under the assumption of aggressive production expansion, the supply of storage chips focused on AI from 2026 to 2030 cannot fill the gap in demand from cloud vendors, a gap equivalent to 450,000 wafers of monthly output; capital expenditure for Iron Man accounts for only 5% of revenue, much lower than the five-year average of over 20%, and Samsung and SK Hynix plan to keep capital spending in the low to mid-single digits over the next two years. On the demand side, with AI servers, agentic AI, enterprise SSDs, and HBM demand accelerating, Iron Man reports a price increase of over 100% every quarter, and DRAM prices have tripled in the past year. SanDisk's performance also confirms this profit elasticity: its Q3 2026 revenue surged 97% quarter-on-quarter and increased by 251% year-on-year to $5.95 billion, with a gross margin of 78.4% and an EPS of $23.41. From the perspective of semiconductor engineering, storage is no longer just a "server accessory" but is now a core constraint on the throughput and economics of AI systems. The training side requires HBM support for high-bandwidth access by GPUs/ASICs, while the inference side requires DRAM and eSSD to support KV cache, long contexts, enhanced retrieval, agent workflow states, and extensive enterprise data retrieval; as the popularity of agentic AI grows, the demand for data read/write, context storage, vector libraries, cache layers, and enterprise-level SSDs also grows stronger. Therefore, storage manufacturers are gaining pricing power for the first time.