AdvertiseFee Deal Desk

Meta Platforms

Samsung Reportedly Building Custom HBM4E for Meta Platforms (META) on 2nm Node

Samsung is reportedly developing custom HBM4E memory for Meta Platforms' in-house AI chips, with the base die made on Samsung's own 2nm foundry process.

Be a creator
November 17, 2026, 12:01 PM UTC4 min read

Samsung's Custom HBM4E for Meta's AI Chips

Samsung Electronics is developing a customized high-bandwidth memory product for Meta Platforms (META), built for the social media company's in-house artificial intelligence chips, according to The Bell, a Korean outlet that cited industry sources for the report. The memory is based on seventh-generation HBM4E, and Samsung plans to manufacture its base die on the 2-nanometer process at its own foundry, a step that would extend a custom-memory customer roster running from NVIDIA (NVDA) to Advanced Micro Devices (AMD). High-bandwidth memory is the stacked DRAM class that sits beside AI accelerators. Bandwidth, not raw capacity, is the metric that matters for accelerator memory, and stacking dies vertically is how HBM achieves it. The base die forms the bottom of every HBM stack, handling the connection between the processor and the DRAM dies above, and memory makers have traditionally kept its design standardized. A custom version changes that: functions specific to the customer's accelerator can be written directly into the base die, which is what Samsung plans for the Meta part. The report adds that Samsung intends to supply standard and custom versions side by side from the HBM4E generation onward, developing custom base dies on 2-nanometer and 4-nanometer processes in parallel to match each customer's requirements. Early custom base dies are expected to integrate intellectual property such as the HBM controller and die-to-die communication interfaces. The die-to-die interface carries data between chips inside a single package. Optimizing those interfaces cuts the power consumed when data moves between memory and processor, a concrete selling point in AI data centers where electricity and cooling have become the primary constraints on expansion. Neither Samsung nor Meta has publicly confirmed the arrangement.

TSMC Enters Samsung's Base Die Equation

Samsung's own foundry is viewed in the industry as its main competitive edge in the custom HBM market, particularly as custom base dies depend increasingly on advanced logic processes. Industry sources say products such as next-generation HBM base dies are gradually filling Samsung's 2-nanometer production lines. In the United States, Maeil Business Newspaper has reported that Samsung plans to produce 2-nanometer HBM base dies at its second fab in Taylor, Texas, with AMD expected to take up most of the planned output. The picture is complicated by NVIDIA, which has asked Samsung, SK Hynix and Micron to build the base die for its NVHBM memory on TSMC's 3-nanometer process, with the aim of unifying production across suppliers and cutting performance gaps and validation work. SK Hynix has already outsourced HBM4 base die production to TSMC, and Micron plans a comparable arrangement from HBM4E, so Samsung must balance its in-house 2-nanometer capability against customer-specified TSMC production. For Taiwan's supply chain, these shifts mean TSMC's role in HBM base dies keeps expanding. Competition is tightening among the memory makers too. SK Hynix confirmed on its third-quarter earnings call that it is working closely with major customers to develop customized HBM projects tailored to their systems and workloads, and that it completed HBM4E sample shipments in the second quarter, with development and customer validation on schedule. Across all of these moves, one pattern holds: as large AI chip developers design their own silicon, they are attaching custom requirements to memory as well, and the base die is becoming the new center of competition in the segment.

The strategic read is that memory, not logic, is becoming the next customization battleground for AI silicon, and a Meta design win would give Samsung a third anchor customer beyond NVIDIA and AMD. For Meta, a custom base die ties its accelerator program to one supplier's memory roadmap, a commitment that runs across product generations. COINOTAG data shows trading in META stock, and its composite support and resistance scoring maps the levels traders are watching around the report.

COINOTAG's editorial and research desk.

AI-Assisted

AI-generated, AI-reviewed, under COINOTAG editorial oversight.