A reported late-September qualification of Samsung’s 12-layer HBM4E with Nvidia and major hyperscalers would be a significant technical milestone in the competition for AI-memory orders. It would suggest the chip met important customer requirements—but it does not establish that Samsung has won large supply contracts or begun mass production. The qualification report has not been confirmed by Samsung, and supply volumes, contract values and a production schedule have not been disclosed.
13
3
Qualification is a hurdle, not a purchase order
Customer qualification can make a product eligible for consideration in a customer’s systems. That is meaningful for Samsung as it seeks a stronger position in high-bandwidth memory (HBM), but a reported pass alone does not show how much memory a customer will buy—or when it will ship. Industry coverage also cautions that a qualification or design win does not automatically translate into an immediate change in market share.
3
The commercial outcome depends on further steps, including customer orders and Samsung’s ability to produce the chip at volume. Until those details are public, the report is best read as evidence of progress, not as proof of a revenue-generating supply win.
3
13
What Samsung has disclosed about the 12-layer HBM4E
Samsung announced shipments of 12-layer HBM4E samples to global customers on May 29. The company listed the stack at 48 GB, with a stable pin speed of 14 Gbps that can scale up to 16 Gbps. Samsung says the speed is more than 20% higher than HBM4, and gives bandwidth of up to 3.6 TB/s for the 12-layer product.
16
33
The chip uses sixth-generation, 10-nanometer-class 1c DRAM and a 4-nanometer foundry logic base die, according to Samsung’s product information.
30
37
One specification needs careful context: 4.0 TB/s is not the published bandwidth for the 12-layer, 48-GB configuration. Samsung identifies that figure for a 16-layer stack with 64 GB of capacity.
16
30
HBM4 and HBM4E have different timelines
Samsung began mass-production shipments of HBM4 in February 2026. It then showed HBM4E at Nvidia’s GTC event in San Jose in March and announced 12-layer HBM4E sample shipments in May. These dates mark progress across two generations of memory; HBM4 production does not mean HBM4E is already in mass production.
17
20
33
The reported late-September qualification would be a later step for the HBM4E samples. However, the available reporting does not provide a confirmed date for HBM4E volume production.
13
3
What to watch next
A separate report described a custom, faster-pin-speed, eight-layer HBM4E aimed at Nvidia. That remains a separate, unconfirmed development and should not be confused with the reported qualification of Samsung’s 12-layer product.
26
The next evidence of a commercial win would be clearer information about orders, supply volumes and production. Until then, qualification may strengthen Samsung’s position as a potential supplier, but the scale and timing of any shift in customer allocations remain uncertain.
3
13