[News] NVIDIA Reportedly Eyes Small-Scale HBM Base Die Production in 2027: Rattling Memory, Chip Markets

Market rumors suggest NVIDIA has kicked off development of its own HBM base die, sending ripples through the supply chain as it could reshape the next-generation HBM landscape. According to the Commercial Times, the chip is expected to be built on the 3nm node, with small-batch trial production scheduled for the second half of 2027.

NVIDIA’s Strategy and HBM4 Roadmap

The move, according to TrendForce, indicates NVIDIA is advancing toward customized HBM base dies that integrate portions of GPU functionality into the base layer, targeting enhanced overall performance for both HBM and GPU systems.

Notably, as TrendForce observes, NVIDIA will first adopt standard HBM4e in the first half of 2027—supplied by SK hynix on TSMC’s 12nm process—before transitioning to customized HBM4e designs from the second half of 2027 through 2028, with SK hynix supporting production on TSMC’s 3nm node.

Industry-Wide Implications: From Memory to ASICs

Commercial Times notes that while SK hynix currently dominates the HBM market with largely in-house base die designs, achieving I/O rates above 10Gbps requires advanced logic nodes, such as TSMC’s 12nm or smaller, for base die manufacturing.

As memory manufacturers lack sophisticated base die IP and ASIC design capabilities, sources cited in the report highlight that integrating UCIe interfaces into HBM4 for direct GPU and CPU connections would significantly increase design complexity, the report adds.

Against this backdrop, Commercial Times suggests NVIDIA’s plan to develop proprietary HBM base dies is seen as a strategic move into the ASIC market—leveraging NVLink Fusion to offer more modular solutions and strengthen ecosystem control.

However, since CSPs have already invested in ASICs to reduce NVIDIA dependence, the move’s actual impact on ASIC vendors may be limited, the report concludes.

In the meantime, industry insiders cited by Commercial Times suggest the biggest beneficiary could be TSMC, whose close collaboration with major chipmakers and continuous process improvements place it at the heart of AI’s rapid advancement.

Read more

• [News] NVIDIA Reportedly Denies Rubin Delay as AMD MI450 Spurs Redesign Speculation

(Photo credit: NVIDIA)