Samsung is expected to ship HBM4 memory chips to Nvidia, Barron's reported. AI accelerator demand is outrunning foundry and memory supply, forcing multi year pre booking and qualification schedules.
Nvidia's compute cycle is shifting scarce semiconductor resources across the stack, from TSMC wafers to HBM supply, which limits how quickly rivals can catch up.
Before this cycle, Apple was widely viewed as TSMC's largest customer and HBM supply was constrained across vendors. When AI driven demand tightens supply of critical components like HBM, suppliers can sell out years ahead. HBM chips carry higher margins than typical memory components and are crucial for AI processors. TSMC is the dominant contract chip manufacturer by market share per CNBC citing TrendForce.
CNBC reported Nvidia is projected to become TSMC's largest customer this year. CNBC cited an estimate that Nvidia could account for about 22% of TSMC revenue versus about 18% for Apple. Barron's noted analysts expect tight HBM supply across Samsung, Micron, and SK Hynix, limiting share shifts. SK hynix and Micron said they were out of HBM for 2024 and 2025 was nearly sold out, per CNBC.
Rising Nvidia chip demand drives greater foundry capacity needs while pulling in high bandwidth memory supply commitments. Samsung starting HBM4 production next month forces rivals to compete for remaining memory allocations. Nvidia at about 22% of TSMC revenue concentrates bargaining power as Apple remains about 18%. Creative Strategies says larger AI accelerators increase packaging intensity, changing how TSMC allocates capacity and prices nodes. Tight supply across Samsung, Micron, and SK Hynix limits quick switching even as demand rises.
HBM sells out years ahead while TSMC customer exposure estimates reorder internal scheduling and prioritization. The memory stack carries higher margins, while the foundry layer concentrates share in leading node customers. Multi year pre booking constrains incremental ramps for late movers and supports supplier pricing power. Analyst commentary says exact HBM4 share will be decided in 2026 while supply stays tight. Greater foundry needs and pulled memory commitments turn volume ramp into a procurement contest.
Actual Samsung volume allocations to Nvidia remain undisclosed. Contract pricing for HBM4 and wafer capacity commitments stays undisclosed. Whether the projected customer ranking at TSMC will be confirmed remains uncertain. TSMC confirmation limits claims about being the largest customer. Supply tightness limits claims about Micron losing Nvidia business. Bill of materials opacity limits estimates of how memory affects output. The question is how 2026 HBM4 share decisions and wafer supply constraints will determine rivals' volume ramps.
