Test report DSG-9841 · Rev B · tested October 10, 2026
Memory & StorageDevice under test
SK Hynix begins testing HBM5 with TSMC as HBM4 enters Nvidia's next platform
SK Hynix has begun HBM5 testing with TSMC while HBM4 enters Nvidia's next platform, running two HBM generations in parallel as AI memory cycles compress.
- Read
- 2 min
- Words
- 359
- Node
- 5nm
- Operator
- Priya Raman
Spec summary
- SK Hynix has started testing HBM5 with TSMC
- HBM4 is entering Nvidia's next platform
- HBM5 foundry-side testing begins before HBM4 reaches full volume
- Samsung and Micron are competing for the same HBM4 customers
SK Hynix has started testing HBM5 memory with TSMC, according to a report from Digitimes, while its current-generation HBM4 modules are entering qualification for Nvidia's next platform.
The overlap marks an unusually early start for next-generation HBM5 validation. High-bandwidth memory suppliers typically begin foundry-side testing of a new HBM generation only after the previous one has completed customer qualification at scale. SK Hynix is now running both tracks in parallel.
What does the HBM5 testing involve?
Testing next-generation HBM with TSMC addresses the integration between advanced memory stacks and the foundry's logic process nodes. HBM modules must be validated for compatibility with the CoWoS-style packaging that TSMC supplies for AI accelerators, including GPUs designed by Nvidia and other major chip developers.
Successful early testing at TSMC can shorten the path from HBM5 silicon to volume shipments once AI accelerator vendors commit to the new standard.
Where does HBM4 stand?
HBM4 is no longer a future product line for SK Hynix. The report states that HBM4 is entering Nvidia's next platform, meaning the memory generation is moving from development and sampling into an active qualification cycle tied to a specific, near-term silicon roadmap.
For SK Hynix, HBM4 placement in an Nvidia platform secures continued participation in the dominant AI accelerator supply chain, where high-bandwidth memory content per GPU continues to rise with each product generation.
Why run both generations at once?
Parallel development reflects the compression of AI memory product cycles. Nvidia's platform cadence now pulls HBM generations forward faster than the historical server-memory replacement cycle, pushing suppliers to qualify the next generation before the current one reaches full volume.
SK Hynix's early HBM5 work at TSMC positions the company to defend its HBM market share against Samsung Electronics and Micron, both of which are racing to supply HBM4 to the same customers.
What comes next?
The report does not specify a timeline for HBM5 mass production or name the specific Nvidia platform receiving HBM4. Watch for qualification milestones at TSMC and volume shipment schedules as indicators of when each generation transitions from testing to revenue.
via Google News: TSMC (Source)
More from Priya Raman
Same lot · LOT-C1C6
- DSG-29493nmSK hynix, Samsung push 16-layer HBM4 as NVIDIA Rubin favors 12-layer
- DSG-460920nmSamsung Passes Nvidia Quality Tests for HBM4E Memory Chips
- DSG-87827nmSamsung Passes HBM4E Qualification with Nvidia and Major Hyperscalers
- DSG-959345nmMicron Targets NVIDIA Custom HBM Standards in TSMC Partnership
- DSG-774020nmNVIDIA's NVHBM Pushes Micron and Samsung Toward TSMC Base Dies