Citi Urges Investors to Buy Memory Dip as AI Infrastructure Cycle Accelerates

Citi is advising investors to "buy the dip" in memory chip stocks following a sharp market correction. The bank argues that the current AI-driven memory cycle is still in its early stages, suggesting that recent price drops represent a buying opportunity rather than a peak in the technology's adoption curve.

Major industry players, including Micron, Samsung, and SK Hynix, have seen their stock prices decline by more than 20% from recent peaks. This sell-off was primarily driven by investor concerns regarding high valuations and the long-term sustainability of AI capital expenditure. Despite this volatility, Citi maintains a bullish outlook, upholding a "Buy" rating for SK Hynix with a price target of 3,100,000 won. The bank further signaled confidence in the company by raising its 2026 and 2027 operating profit forecasts by 4% and 3%, respectively.

The Shift to Structural Growth

The memory chip market has historically been defined by volatile boom-and-bust cycles. However, Citi suggests the current era differs from previous surges due to a fundamental shift in how memory is procured and deployed. A key driver is the rise of long-term agreements (LTAs) spanning three to five years. These contracts provide manufacturers with substantial advance payments and significantly improve earnings visibility compared to the spot-market reliance of the past.

Citi analysts note that as mid-to-long-term earnings visibility clears, supported by these LTAs and the ongoing AI upcycle, companies like SK Hynix are positioned to provide a constructive market outlook and decent shareholder returns. This transition suggests that the industry is moving away from the erratic pricing swings that characterized previous decades of semiconductor manufacturing.

Infrastructure Implications

The scale of the projected infrastructure build-out is massive. Citi predicts that total High-Bandwidth Memory (HBM) capacity per AI system will increase by 434%, growing from 20.7 terabytes to 110.6 terabytes. This growth is expected to be driven by a rise in GPU counts per system, which the bank projects will climb from 72 to 576.

This projection indicates that the investment thesis for memory stocks is shifting from short-term speculation on GPU hype to a long-term structural growth play. If these capacity increases materialize, the benefit to memory manufacturers will extend far beyond the initial wave of AI hardware deployment, creating a durable demand floor for HBM that protects against sudden market pivots.

Market Outlook

While the market remains sensitive to AI spending levels, Citi views the current correction as temporary. The bank's analysis suggests that the integration of HBM into AI systems is creating a more resilient growth phase than the industry's previous major booms.

Investors will now be watching to see if the projected increase in GPU density per system manifests in hardware orders and whether other major memory producers follow SK Hynix's lead in securing multi-year LTAs to stabilize future revenue. The ability of the sector to maintain these high-capacity trajectories will likely determine if the current dip is indeed the floor for the next decade of AI expansion.