TrendForce's Memory Mega-Upgrade: Mapping the Winners in a USD 1.28 Trillion Market

TrendForce significantly raised its global memory market forecast this week, projecting the sector will exceed USD 1.28 trillion by 2027. The driver is straightforward: AI infrastructure buildout has created a structural step-change in demand for High Bandwidth Memory (HBM) and advanced NAND that conventional memory forecasting models did not anticipate at this scale.

To put the number in context: the global memory market bottomed at roughly USD 180 billion during the 2023 downcycle. A USD 1.28 trillion forecast by 2027 implies a roughly 7x expansion from trough. Even applying meaningful analyst optimism discount, the directional case for sustained memory market growth is supported by visible demand from hyperscaler AI capex commitments.

HBM: The Tier-1 Beneficiary

HBM is the memory architecture that sits directly on Nvidia AI accelerator packages. Each H100 uses approximately 80 GB of HBM3E. The Rubin generation, announced today by Jensen Huang, will require even higher HBM density per chip to support its 5x inference performance improvement.

SK Hynix holds the largest HBM market share today and was first to qualify HBM3E with Nvidia. Samsung is ramping HBM3E but has faced yield challenges. Micron (MU.US) qualifiedwithNvidiamorerecentlyandisexpandingproductioncapacityaggressively.SanDisk(MU.US)qualifiedwithNvidiamorerecentlyandisexpandingproductioncapacityaggressively.SanDisk(SNDK.US) is positioning its upcoming High Bandwidth Flash (HBF) as a complementary inference-layer solution, targeting a different part of the AI compute stack.

NAND: The Second-Order Beneficiary

AI training clusters require massive NVMe SSD storage arrays for dataset management. AI inference requires fast retrieval of model weights and KV-cache data. Both applications drive enterprise SSD demand at a scale that enterprise NAND suppliers including SanDisk, Samsung, and Kioxia are positioned to capture.

SanDisk's HBF announcement, co-standardised with SK Hynix under the Open Compute Project, is a direct attempt to position flash storage as the next layer of the AI memory hierarchy after HBM.

Key Risk

Forecasts of this magnitude have historically been revised downward when AI capex cycles moderate. If hyperscaler spending plateaus in 2026, the demand curve could flatten before reaching the USD 1.28 trillion target. Inventory cycles in memory markets can reverse quickly. Investors should track quarterly HBM order patterns and AI server shipment data as the most reliable leading indicators.