Nvidia’s NVHBM Could Shift Memory Pricing Power, Analyst Argues
Shared specifications could make qualified suppliers easier to substitute, but any change in bargaining power depends on available production capacity.

Nvidia’s NVHBM design could give the chipmaker more leverage over high-bandwidth memory suppliers if common specifications make qualified products easier to substitute, semiconductor analyst Damnang argues.
Nvidia introduced NVHBM in August 2026 as a custom product based on HBM4E. It set shared manufacturing specifications for Samsung Electronics, SK hynix and Micron to support consistent performance and reduce the work required to validate each supplier’s chips.
A high-bandwidth memory stack has a logic base die beneath the DRAM dies that store data. The base die handles control functions. Damnang argues that a common design could give Nvidia greater influence over the base die and system integration, leaving memory makers to compete more directly on the DRAM they produce.
That pricing argument depends on more than a shared specification. Another supplier would have to qualify its product and have enough capacity to fill orders. If those conditions are met, Nvidia could shift purchases more easily, putting pressure on suppliers during contract negotiations. Producers that improve manufacturing yields could also have less room to retain the resulting savings.
The ability to switch suppliers remains a condition for that outcome, rather than an established result of the design. Each supplier’s chips still need to meet Nvidia’s requirements, and available output would determine how much of an order another producer could take. Those practical limits matter to any change in bargaining power.
Samsung plans to use Taiwan Semiconductor Manufacturing Co.’s 3-nanometer process for its NVHBM base dies. Its standard HBM4 and HBM4E base dies use Samsung’s own 4-nanometer process. The NVHBM plan brings an outside foundry into the production of the logic component while Samsung makes the dies that store data.
AWS’s Annapurna Labs is also working with Nvidia and memory makers on NVHBM. That cooperation adds another participant to the effort, though it does not establish a use for NVHBM in a particular Annapurna processor.
Separately, Samsung has proposed a price in the mid-$4 range per gigabit for 2027 HBM4, compared with roughly $1.50 per gigabit for HBM3E. Those figures concern HBM4 and HBM3E, not NVHBM, which is based on HBM4E; they do not establish an NVHBM price. Whether the shared design changes future contract terms will depend on supplier qualification and capacity.