Nvidia reportedly wants all its HBM suppliers to use TSMC’s 3nm process
Nvidia requires suppliers (Samsung, SK Hynix, Micron) to use TSMC’s 3nm process for NVHBM base dies, aiming for consistency. Samsung agreed, marking their first collaboration on HBM. NVHBM offers 30% more bandwidth, 15% less power than standard HBM4E. UBS forecasts a 10% HBM supply gap in 2024. TSMC plans 3nm production expansion. Samsung maintains turnkey approach for other HBM products.
How this was made
The 30-second read
Why it matters
The coordination could streamline validation, reduce design variance, and accelerate AI hardware rollouts, but also creates reliance on TSMC's capacity timeline.
Market read
The announcement could influence AI hardware supply chains and affect stock sentiment for Nvidia and its memory suppliers.
What to watch
Potential yield issues on early 3nm runs and the cost premium of TSMC versus Samsung's own processes.
Background
Nvidia's NVHBM custom memory aims to improve AI GPU efficiency. The shift to TSMC's 3nm process aligns all three suppliers—Samsung, SK Hynix, Micron—on a common node.
Ticker impact
Nvidia disclosed it wants all NVHBM suppliers to produce the base die on TSMC's 3nm process, a first-time partnership that could improve GPU performance.
likely upward pressure as investors price in performance gains from the new 3nm HBM base die.
The move reduces validation costs and promises up to 30% more bandwidth, which is material for Nvidia's product roadmap.
Micron, a listed NVHBM supplier, is required to shift its base die production to TSMC's 3nm process, moving away from its current approach.
possible slight downside as Micron incurs transition costs and loses control of its own base‑die roadmap.
The requirement forces Micron to rely on TSMC capacity, which could affect margins until volume ramps.
Market effects
Highlights a shift toward advanced-node foundry services for high‑bandwidth memory, benefiting TSMC and potentially pressuring other memory fabs.
May boost Taiwan's semiconductor outlook while creating capacity concerns for South Korean foundries.
Signals tighter supply dynamics for AI‑driven GPUs, a key theme in the broader tech sector.
Counterpoint
If TSMC capacity constraints delay mass production, Nvidia's performance gains could be postponed, hurting the hype.
Key entities
- CompanyNvidia
AI chipmaker seeking performance gains via new HBM process.
- CompanyTSMC
Foundry providing 3nm process for HBM base dies.
- CompanyMicron
US-listed memory supplier affected by the new requirement.
