Micron says most of its 2027 HBM is contracted and confirms a foundry-made base die for NVIDIA's NVHBM
UPDATEDMicron's fiscal Q4 remarks say the vast majority of its calendar 2027 HBM supply is under agreement, with 26 take-or-pay customer agreements covering over 35% of revenue through 2030. Micron also confirmed it is co-designing NVIDIA's custom NVHBM, with base dies made on an external foundry process. RTX PRO 5500 is still 'Coming Soon', and the reported Chinese approval is still unconfirmed. H200, HGX B200 and RTX PRO 6000 are unchanged on vendor pages.
NVDA ecosystem — weekly dispatch (2026-W41)
### Memory: Micron's HBM for 2027 is largely committed
In its fiscal Q4 2026 prepared remarks (30 September), Micron said it has completed agreements for the vast majority of its calendar 2027 HBM bit supply, at significantly higher prices than in 2026. It has signed 26 take-or-pay Strategic Customer Agreements, which it estimates at over 35% of revenue through 2030, and some now run into 2031. Micron also said it has no line of sight to DRAM supply and demand returning to balance. Micron's HBM advanced-packaging facility in Singapore is ahead of plan, with initial output in early 2027. ID1 in Idaho remains on track for wafer output in mid-2027. This is recorded as an event under Supply chain.
### NVHBM: a foundry-made base die
Micron confirmed it is co-designing NVIDIA's custom HBM4E, NVHBM, for the next GPU generation and NVLink Fusion platforms. On the call, CTPO Scott DeBoer said NVHBM and Micron's standard HBM4E both use base dies made on a foundry process. Micron did not name the foundry. The Elec reports it is TSMC, which is unconfirmed. NVHBM moves NVIDIA's memory controller into the base die, so that die is now logic silicon. Samsung's HBM page already describes its HBM4E on a 4 nm foundry base die. This is recorded as an event under Chip journey and Supply chain.
### RTX PRO 5500: still "Coming Soon"
The vendor page is unchanged from W40: 84 GB GDDR7 ECC, 1,398 GB/s, up to 600 W, two-way MIG, rack-mounted, availability "Coming Soon". The reported MIIT approval for ByteDance and Alibaba purchases has still not been confirmed by MIIT, NVIDIA or the buyers.
### Regulation and enforcement
The US remote-access rule logged in August still has no published text. On 1 October, the Justice Department charged a California company owner with smuggling more than $300 million of export-controlled servers containing NVIDIA chips to China through Malaysia and Singapore in 2023–24. This is enforcement of existing rules, not a new rule, so it is not logged as an event.
### Catalog re-verification
- H200 SXM: 141 GB, 4.8 TB/s, up to 700 W configurable, 3,958 TFLOPS FP8. Unchanged.
- HGX B200: 1.4 TB total memory, 1.8 TB/s NVLink per GPU, 14.4 TB/s total, 72 PFLOPS FP8. Unchanged. The page lists HGX B300 and B200 as "shipping now" and Rubin as "ramping into full production".
- RTX PRO 6000 Blackwell: 96 GB GDDR7, 1,792 GB/s, 600 W. Unchanged.
Catalog unchanged.
---
Technical research for informational purposes only. Not financial advice. No investment recommendations. Specs sourced from vendor documentation; verify before engineering or procurement decisions.
NVHBM changes how the HBM stack is built: the base die becomes foundry logic silicon carrying NVIDIA's memory controller. That puts advanced-node foundry capacity into the memory path as well as the GPU die, at a time when Micron says it has no line of sight to DRAM supply balance.