BofA: DRAM Prices to Rise 20-30% QoQ in Q3, Cloud Vendors Locking in Price Contracts Through 2027
nashnova research
BofA Securities confirms Q3 DRAM prices rose 20-30% QoQ while hyperscale cloud buyers have already signed contracts locking in higher DRAM prices for Q1 2027 — buyers voluntarily accepting price hikes in writing, signaling the memory upcycle has entered a new phase.
How much did prices rise, and who's paying up?
BofA analyst Simon Woo published a Sept 19 note: Q3 DRAM ASPs rose 20-30% QoQ; NAND rose over 15%.
The bigger signal: hyperscale cloud companies have already signed contracts locking in higher DRAM prices for Q1 2027. This means → buyers aren't passively accepting increases — they're voluntarily writing price hikes into contracts, betting the supply shortage will persist.
Q4 won't be weak either. BofA expects at least single-digit QoQ ASP gains.
Why is supply so tight?
The sufficiency ratio — supply ÷ demand, where anything below 100% means shortage — is the key metric. 2026 DRAM sufficiency is only ~79%, far below the 100% balance line; NAND sits at roughly 80%.
In plain terms = for every 100 memory chips the market needs, suppliers can only deliver about 80. The gap is obvious.
Sufficiency won't recover toward equilibrium until 2027, when NAND reaches roughly 102%. This reflects a structural capacity shortfall, not short-term speculation.
How powerful is the AI pull?
AI server demand for HBM — high-bandwidth memory, an ultra-fast memory designed specifically for AI chips — now accounts for over 50% of total DRAM shipments.
BofA projects the 2026 HBM market at $77.4 billion, up 124% YoY; by 2027 it expands to $152.8 billion. This means → HBM nearly doubles and then doubles again within two years. AI's appetite for memory is nowhere near satiated.
PC and smartphone shipments fell only about 10% YoY — end-user demand hasn't collapsed. The supply-demand gap is driven almost entirely by AI.
Where do prices go in 2027-2028?
BofA raised its 2027-2028 ASP forecasts: DRAM assumptions up 8-12%, now ~$17.1 per 8Gb-equivalent in 2027 and ~$16.2 in 2028; NAND up 2-3%.
BofA expects 2028 ASPs to fall ~10% from 2027 (DRAM down 5%, NAND down 13%), which it characterizes as a "soft landing." In plain terms = prices won't cliff-drop — they'll ease down gradually.
Looking further out, BofA sees 2029-2030 driven by next-generation AI chips and advanced memory solutions, sparking a fresh growth leg.
What supports the $2 trillion long-term target?
BofA raised its 2030 global memory market (DRAM + NAND) forecast from $1.8 trillion to $2.0 trillion, projecting a 21% CAGR from 2027 to 2030.
Five supporting reasons: Q3 2026 annualized run-rate already exceeds $1.0 trillion; 2026 ASPs are up 3-4× YoY with volume growth over 20%; record-high ASPs look sustainable; annual bit growth stays in the high double digits; advanced solutions like HBM and eSSD (enterprise solid-state drives) keep gaining share.
BofA's proprietary "Memory Boom Index" hit a record 180 in July — the cycle midpoint is 100, and the 2017-2018 peak reached only 120. This reflects a cycle whose intensity already dwarfs any historical comparable.
Where is the risk in this super-cycle?
Hyperscalers voluntarily locking in 2027 price-hike contracts shows demand-side consensus on high prices — but consensus also means that if demand undershoots, committed contracts will amplify inventory pressure.
Whether 2028 delivers a soft landing or a hard one is the critical test for this super-cycle's durability. This means → a modest price decline is healthy, but if the drop exceeds BofA's projected 10%, market confidence could reverse quickly.
Korean semiconductor exports and monthly sales at Taiwan's Nanya Technology and Phison Electronics are all up over 100% YoY — near-term momentum remains strong.
市场有风险，内容仅供研究参考，不构成投资建议。