NVIDIA originally requested 9Gbps bandwidth for HBM4, with Samsung failing to hit 9Gbps with its 1b DRAM-based HBM4, but using its 1c DRAM and SF4 to HBM4, they could hit over 10Gbps... but SK hynix needed more voltage, and Micron said 10Gbps was impossible.
In a new post on X by leaker @Jukanrosleve, we're hearing that NVIDIA originally requested 9Gbps for HBM4, but once discovering that 10Gbps was achievable, NVIDIA then asked "why not try 11Gbps?" Jukan says that NVIDIA is known for making such demanding requests all the time, adding that in the long-term, he thinks the spec "will likely end up being set at 10Gbps".
This information and post on X was in response to another post the leaker did about half an hour prior, which reads: "NVIDIA has requested Samsung and SK hynix to raise HBM4 speed to 11Gbps, an additional increase from the previous 9Gbps to 10Gbps target".
- Read more: NVIDIA to Samsung: 'we'll double' GDDR7 order, 'so be ready'
- Read more: Samsung HBM4 passes NVIDIA's tests, mass production now imminent
- Read more: Samsung confident for semiconductor business in 2026: starts with HBM4
- Read more: Samsung's next-gen 1c DRAM test yields for its HBM4 rumored at 65%
What's interesting here is that Samsung was failing in its push with HBM3 and HBM3E, putting all of its semiconductor future into a major overhaul of its semiconductor division, and its eggs into the next-gen HBM4 memory basket, which seems to be paying off.
SK hynix has been dominant in supplying masses of HBM3 and HBM3E memory chips to NVIDIA for use in its mega-successful AI GPUs, but Samsung hopes to be the leading HBM4 memory supplier to NVIDIA in 2026 and beyond.




