The memory chip shortage shows no signs of easing, with cloud giants already securing contracts at elevated prices for the coming year.

On September 19, BofA Securities analyst Simon Woo released his latest memory industry report, revealing that channel checks confirm DRAM average selling prices (ASPs) will climb 20-30% quarter-on-quarter in Q3 2026, while NAND prices rise over 15%. Notably, hyperscale cloud providers have already signed new agreements locking in higher DRAM prices for Q1 2027.

Given the persistent supply shortfall and intensified chip procurement by OEMs and tech companies, BofA has raised its DRAM ASP forecasts for 2027-2028 by 8-12% and NAND by 2-3%. However, the firm still expects prices to retreat roughly 10% from 2027 peaks in 2028, describing this as a "soft landing."

On a longer horizon, BofA has lifted its 2030 global memory total addressable market (TAM) projection from $1.8 trillion to $2.0 trillion, while forecasting a 21% compound annual growth rate (CAGR) for combined DRAM+NAND sales between 2027 and 2030.

Q3 pricing exceeds expectations as hyperscalers pre-commit to next year's increases

Channel checks confirm robust memory pricing in the third quarter. The BofA report states: "Our latest checks with memory makers, spot market distributors, and OEMs confirm strong Q3 ASPs: despite a rising share of long-term agreements, most DRAM products saw ASPs climb 20-30% quarter-on-quarter, with NAND rising over 15%."

Q4 momentum remains solid as well. The analyst notes, "Q4 ASPs are expected to grow at least in the high single digits."

More striking is a new development: "Hyperscalers have signed new contracts agreeing to pay higher DRAM prices in Q1 2027 than in Q4 2026," the analyst writes. This indicates buyers are proactively accepting price hikes and embedding those expectations into contractual terms.

2027-2028 forecasts raised, but 2028 points to a soft landing

Based on the supply-demand dynamics, BofA has revised its earlier projections. The analyst explains: "Given the ongoing memory supply shortage and more aggressive chip procurement by certain OEMs and large tech firms, our outlook for H1 2027 has turned more optimistic."

Specifically:

DRAM: ASP assumptions for 2027-2028 raised by 8-12%; in the new forecast, 2027 ASP lands at approximately $17.1 per 8Gb equivalent, with 2028 at roughly $16.2.

NAND: ASP assumptions for 2027-2028 raised by 2-3%; 2027 at approximately $10.0 per 256Gb equivalent, and 2028 at around $8.7.

Meanwhile, BofA holds its H2 2026 estimates unchanged—"having already held an optimistic view on H2 ASPs, no adjustments are made this time."

The analyst also points out, "2028 ASPs are expected to decline approximately 10% from 2027 (DRAM -5%, NAND -13%)," framing this as "a soft landing phase ahead of renewed growth in 2029-2030, driven by more advanced AI and GPU/ASIC platforms utilizing high-end memory."

Supply constraints remain the core thesis

Why can prices keep climbing? Supply-side constraints are pivotal. Data shows DRAM supply-to-demand sufficiency in 2026 sits at just 79%, well below the 100% equilibrium line, only gradually recovering toward balance by 2027. NAND follows a similar pattern, with sufficiency around 80% in 2026 before recovering to roughly 102% in 2027.

Production cuts in PC and smartphones are also justified. The report notes that despite the memory shortage, "year-to-date production cuts in PC and smartphones are not severe, with shipments declining only about 10% year-on-year." This means end-demand has not collapsed; the supply-demand gap is primarily driven by AI servers' robust appetite for memory.

On the server side, AI servers' demand for HBM (High Bandwidth Memory) now accounts for over 50% of total DRAM shipments. The analyst projects the HBM market to reach $77.4 billion in 2026, up 124% year-on-year, expanding further to $152.8 billion in 2027.

2030 TAM target raised to $2 trillion, cementing a long-term supercycle

BofA has upgraded its 2030 global memory market (DRAM+NAND) TAM forecast from $1.8 trillion to $2.0 trillion. The analyst notes that global DRAM+NAND sales previously peaked at $0.17 trillion in 2018 (during the cloud boom), while the 2025 recovery generated only $0.21 trillion (amid a NAND downturn). Thus, investors may question the $2.0 trillion projection, given it represents a 10x increase over 2018/25 levels.

Five reasons underpin this forecast:

(1) As of Q3 2026, annualized run-rate already exceeds $1.0 trillion ($278 billion; Q4 equivalent: $1.1 trillion);

(2) 2026 ASPs or contract prices rise 3-4x year-on-year, with bit shipments growing over 20%;

(3) Record-high ASPs are sustainable, with limited price erosion through 2030 due to tight global supply and robust AI chip demand;

(4) Annual bit growth should remain in the high double digits from 2026-2030;

(5) The mix should improve as advanced AI solutions (HBM, SOCAMM, eSSD, etc.) gain share.

The analyst also does not anticipate "down-binning" or lower memory content in new GPU/ASIC designs; instead, assumptions point to further increases—Rubin Ultra at 1,000GB per unit, new Rubin at 384GB, versus current 288GB.

Looking ahead, BofA projects combined DRAM+NAND sales to grow at a 21% CAGR from 2027 through 2030. This trajectory implies DRAM+NAND sales reaching $1.4 trillion by 2030, up from $0.9 trillion in 2026—clearly signaling a prolonged supercycle. This also implies a reasonable market cap of $10 trillion for the memory industry, based on 13x P/E, 50% operating margin on $2.0 trillion sales, and a 25% tax rate.

Additionally, BofA's proprietary "Memory Indicator," which tracks year-on-year momentum across ASPs, revenue, spot prices, and exports, remains instructive. The analyst notes the index "held at a record 180 in July, versus a cycle median of 100 and a prior peak of 120 in 2017-18." BofA expects August-September data to stay strong, with Korean semiconductor exports and monthly sales from Taiwan's Nanya Technology and Phison Electronics all showing year-on-year growth exceeding 100%.