Now Playing
Memory Controller Inside the Stack: NVHBM, NVLink Lock-In & Samsung HBM5
Audio is available on Spreaker — see link below.
Nvidia's NVHBM moves the memory controller into the HBM stack itself — unlocking 25% more compute density and deepening hyperscaler lock-in. Today's briefing also covers Samsung HBM5, LG's Samsung Foundry bet, Intel Xeon 6 in regulated AI, and the UALink race against NVLink.
Audio is available on Spreaker — see link below.
Nvidia just moved the memory controller inside the HBM stack itself. That single architectural decision is the story this briefing is built around, because the implications run well beyond memory bandwidth.
Here's what matters strategically. NVHBM doesn't just improve performance numbers.
The MediaTek partnership formalised on August thirty-first adds another layer to this picture. Three pillars: NVHBM infrastructure integration, RTX Spark consumer AI laptops with six OEM partners targeting fall twenty-twenty-six, and automotive AI positioning.
Away from Nvidia, the other major development is a supply-chain realignment driven by Korean industrial policy. LG has publicly committed to Samsung Foundry for a government-backed AI home chip program, bypassing TSMC.
Intel's story this cycle is quieter but concrete. Kasm Technologies expanded its deployment on Intel Xeon Six using the OpenVINO stack.
Samsung's HBM5 roadmap disclosure rounds out the picture on memory competition. HBM5 doubles performance versus HBM4E with a twenty percent efficiency gain and twenty percent reduction in thermal resistance.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.