Morgan Stanley: The AI-driven storage "super cycle" has arrived

Author: Jukan

Compiled by: TechFlow

Morgan Stanley指出，由人工智能（AI）驱动的新一轮存储器“超级周期”已经到来，其强度和逻辑与以往任何周期完全不同。

Morgan Stanley表示，与过去不同的是，this cycle is led by AI data centers and cloud service providers, customers who are less price-sensitive, with inference workloads becoming the primary driver of mainstream memory demand. Meanwhile, recent channel checks show that server DRAM contract prices in Q4 have surged nearly 70%, while NAND contract prices rose 20% to 30%, giving suppliers unprecedented pricing power.

The firm maintains its Overweight ratings on SK Hynix and Samsung Electronics, expecting rising memory prices to push stock prices to new highs, with memory makers' profits significantly exceeding expectations.

Morgan Stanley noted that the core drivers of this cycle have fundamentally changed in nature. Buyers are no longer price-sensitive traditional customers, but AI data centers and cloud service giants engaged in a computing infrastructure arms race.

For these companies, securing memory supply has become a strategic "necessity," reducing price sensitivity to historic lows. At the same time, HBM (High Bandwidth Memory) production is structurally eroding capacity for traditional DRAM. Morgan Stanley emphasized in its report:

"What's special is that current memory demand has evolved into a competition dominated by AI data centers (compute-intensive platforms) and cloud service providers—players far less sensitive to price than traditional customers... The exponential growth in inference demand provides a solid foundation, which fundamentally differentiates this cycle from any previous one."

According to Morgan Stanley’s latest channel checks, the price outlook for DRAM turned sharply bullish within just two weeks. Q4 server RDIMM contract prices jumped approximately 70%, far exceeding earlier expectations of 30%. Spot prices for DDR5 (16Gb) soared from $7.50 in September to $20.90 currently, an increase of 336%. Quoted prices for DDR4 also rose by 50%. Most contract deals are expected to be finalized by month-end, but customer acceptance appears inevitable—driven by fears of further price hikes and supply shortages.

NAND has become a critical component in AI computing infrastructure and video storage. To address capacity constraints, 3D NAND wafer (TLC and QLC) prices are expected to rise 65%–70% quarter-on-quarter. Nearline storage specifications are shifting from 128TB to 256TB QLC SSDs. According to TrendForce, enterprise SSD bit-based server demand will grow about 50% year-over-year by 2026. Samsung’s bit output will be constrained in the first half of 2025 due to its transition from V6 176-layer to V8 321-layer NAND, with only gradual capacity increases in the second half, limiting its annual bit shipment growth to 10%.

Markets are often influenced by "peak fear," assuming that once stocks hit new highs, a reversal follows. However, Morgan Stanley stresses that in this AI-driven market, the ultimate determinant is earnings growth, not historical valuations:

Currently, server DRAM prices stand at $1/Gb, compared to $1.25/Gb at the peak of the cloud supercycle in Q1 2018. Given the scale of AI infrastructure investment and the dynamic demand from hyperscale customers, this cycle’s price peak is highly likely to surpass historical highs. Memory cycles typically last 4 to 6 quarters, with profits gradually materializing. But the key lies in comparison with market expectations—the market shows significantly greater enthusiasm for mainstream memory prices. Valuation does not predict future returns; it reflects supply-demand dynamics, not historical precedents.

Bolstered by strong long-term AI-driven fundamentals, memory price increases have entered 'uncharted territory,' with profit prospects far exceeding broad market expectations. This implies substantial upside potential for stock prices.

"As AI-related capital expenditures continue to expand, memory’s share of total spending may keep rising—pushing price-to-book (P/B) ratios well beyond historical peaks. We believe this is a story of valuation expansion layered on top of cyclical earnings recovery…

We believe analyst estimate revisions are always lagging—for SK Hynix and Samsung, our earnings forecasts for 2026 and 2027 are 31%–48% and 38%–51% above consensus, respectively."

In summary, the drivers behind this memory “super cycle” are more durable, price gains have already exceeded historical records, and profit outlooks are even more optimistic. Combined with strong cyclical performance, this creates a rare investment opportunity for memory manufacturers with pricing power.