Nvidia reportedly wants all its HBM suppliers to use TSMC’s 3nm process

The Seoul Economic Daily reported, citing industry sources, that Samsung Electronics will make the base die—the logic layer—for Nvidia custom high-bandwidth memory chips using TSMC’s 3-nanometre process. If completed, it would mark the first collaboration between Samsung and Nvidia on HBM production.

Analysis

The key distinction is that the reported work concerns the HBM base die, not necessarily the DRAM stacks themselves. If confirmed and qualified, a second source for this logic layer could give NVIDIA more design and sourcing flexibility and improve its leverage in HBM negotiations; it does not, by itself, demonstrate additional HBM supply or remove advanced-packaging constraints. Samsung Electronics may regain a foothold in NVIDIA’s memory supply chain, while TSMC could benefit from wafer work even as Samsung gains relevance. The near-term read-through for NVDA and TSM is therefore modest and conditional, not a basis for revising earnings absent volume, timing, and yield details.

The immediate catalyst is confirmation from the companies; over the next 1–3 months, watch for qualification milestones and identification of the HBM generation and production volumes. Over 6–18 months, successful multi-sourcing could strengthen NVIDIA’s bargaining position and make custom memory integration more strategically important. Key risks are that the report is inaccurate, qualification or yields disappoint, or packaging—not base-die supply—remains the binding constraint. The contrarian point: this is easy to misread as a Samsung win at TSMC’s expense, but the reported arrangement may instead allocate different value-chain layers to each. No near-term material earnings impact is established.

Key Decisions for Investors

• Do not chase NVDA or TSM on the report alone. Treat it as a modest optionality signal until NVIDIA, Samsung Electronics, or TSMC confirms the arrangement and discloses scope.
• Watch for confirmation of the HBM generation, production volumes, qualification status, and who owns design and yield responsibility. Without these details, do not infer a change in NVIDIA’s HBM supply or TSMC revenue outlook.
• If confirmed, assess NVDA’s supply resilience and negotiating leverage separately from total HBM availability; the thesis weakens if packaging capacity remains the production bottleneck or qualification slips.
• Falsifiers: a company denial, failure to qualify the base die on schedule, or evidence that the work is limited to a small program with no meaningful production volume. Until then, prefer monitoring over an event-driven position.