The narrative on Wall Street regarding AI hardware is shifting from a singular focus on "computing power" to a broader emphasis on "storage power." On January 14, Bernstein analyst Mark Newman and his team stated in a newly released report that the AI-driven data explosion is creating an "unprecedented storage super cycle" and dramatically raised their price target for SanDisk Corp. (SNDK) from $300 to $580. For investors, this report provides a crucial piece of the macroeconomic puzzle concerning the shifting bottlenecks in AI infrastructure. However, a bearish report previously issued by J.P. Morgan also serves as a reminder that the specter of cyclicality still looms. "Context" is becoming the new bottleneck. Bernstein's core thesis is built upon the technological path revealed in Nvidia's CES keynote. As large language models evolve towards longer context and multi-turn conversations, the capacity of traditional HBM has become a bottleneck. The report sharply points out that Nvidia's Vera Rubin architecture formally elevates rack-scale SSDs to a critical path for AI inference, transforming SSDs from mere cold storage into an "active context layer" that helps improve GPU utilization. "'Context is the new bottleneck, and memory must be re-architected'—we believe this statement by Jensen Huang at CES is a game-changer for the NAND industry. Our analysis suggests this could increase the NAND demand per GPU by 5x." Amid an explosion on the demand side, the supply side remains exceptionally restrained. Bernstein emphasizes that, aside from YMTC, there is almost no new NAND capacity being added across the entire industry. This extreme supply-demand mismatch is driving up prices, with the average selling prices for NAND and DRAM experiencing a sharp ascent. The report contends that as long as supply remains disciplined, this high-price environment will continue to support manufacturers' profits. "AI model training and inference workloads, richer content creation, and longer data retention requirements are collectively driving this data explosion, creating an insatiable demand for storage and memory that is also price-insensitive, triggering an unprecedented price surge." However, just as Bernstein is championing the "super cycle," J.P. Morgan recently offered a dose of cold water. In its December report, J.P. Morgan assigned SanDisk only a "Neutral" rating with a price target of $235—less than half of Bernstein's target. J.P. Morgan analyst Harlan Sur believes that SanDisk's current high profits are more indicative of a "cyclical boom" rather than a structural improvement. Given SanDisk's modest 2%-3% market share in the crucial AI enterprise SSD segment, where it holds a follower position, its long-term profitability is seen as unstable. "Short-term excess profits are unlikely to support long-term valuation expansion, potentially leading to a return to the historical 'boom-bust' cycle pattern. As major manufacturers initiate a new round of capacity expansion around 2027, the supply-demand dynamics are expected to loosen." This represents a classic debate over whether "this time is different." Bernstein is betting that AI has fundamentally reshaped the demand function for storage, arguing the market underestimates the value of "context memory." Conversely, J.P. Morgan adheres to the principles of cyclicality, warning that mean reversion is inevitable once new capacity comes online. For investors, the chasm between the $580 and $235 price targets for SanDisk precisely illustrates the vast divergence in current market perceptions regarding the sustainability of the AI hardware boom.
Comments