Samsung Electronics will reportedly use TSMC’s 3-nanometre process to manufacture base dies for Nvidia’s next-generation custom high-bandwidth memory chips, according to the Seoul Economic Daily and Reuters. If confirmed, the arrangement would mark the first time Samsung has worked with its Taiwanese foundry rival at this stage of HBM manufacturing. The move reflects how complex AI-chip supply chains have become as Nvidia pushes suppliers toward customized memory architectures.
What a base die does
High-bandwidth memory is built by stacking multiple memory dies vertically. A base die sits beneath the memory stack and handles important interface, logic and connectivity functions.
As HBM becomes more customized, the base die is increasingly important for performance, power efficiency and integration with GPUs or AI accelerators.
Why TSMC’s 3nm process matters
TSMC’s advanced process nodes are widely used for high-performance logic chips. Moving a base die to 3nm can provide more transistor density and lower power consumption than older processes.
For AI systems operating at enormous scale, even modest efficiency improvements can translate into substantial power and cooling savings.
Samsung and TSMC are usually competitors
Samsung operates both a memory business and a semiconductor foundry business, while TSMC is the world’s largest pure-play contract chipmaker. The two companies compete for advanced manufacturing customers.
A cross-company production arrangement therefore shows that the AI supply chain can force unusual partnerships when one supplier has the required technology or capacity.
Nvidia is reshaping the memory market
Nvidia’s GPUs have become central to the AI infrastructure boom. Demand for HBM has increased rapidly because the memory provides far more bandwidth than conventional DRAM.
AI accelerators need to move huge volumes of data quickly between compute cores and memory, making HBM a critical performance component.
Custom HBM is the next step
Traditional HBM products are relatively standardized, but large AI customers increasingly want customized designs optimized for specific processors and workloads.
That pushes memory manufacturers toward closer integration with foundries, packaging companies and chip designers.
Samsung is trying to strengthen its HBM position
Samsung has faced intense competition from SK Hynix and Micron in advanced HBM. Winning more Nvidia business is strategically important because Nvidia remains the largest customer in the AI accelerator market.
Using TSMC for part of the product could allow Samsung to focus its own resources on memory stacking and packaging while accessing a leading logic process.
Packaging is becoming as important as transistor scaling
Advanced AI chips are no longer defined by a single piece of silicon. They combine GPUs, memory, interconnects and packaging technologies into complex systems.
That means foundry capacity, advanced packaging and HBM supply must all scale together.
Supply-chain interdependence is increasing
A single Nvidia platform can involve design work in the United States, logic manufacturing in Taiwan, memory production in South Korea and packaging across several Asian facilities.
This global interdependence also creates geopolitical and logistical vulnerabilities.
What investors will watch
The key questions are whether the reported partnership becomes a commercial production program, which Nvidia generation it will serve and how much volume Samsung can win.
Any meaningful contract could affect competitive expectations for Samsung, SK Hynix, Micron and TSMC.
A sign of the AI industry’s new structure
The report shows that even the world’s largest semiconductor companies increasingly depend on one another. AI hardware is becoming too complex for one company to dominate every manufacturing stage.
That reality may produce more partnerships between traditional rivals as demand for high-performance memory and advanced logic continues to grow.
Sources
- Seoul Economic Daily report
- Reuters, Oct. 18, 2026
HBM competition is now a strategic contest
SK Hynix established an early lead in supplying advanced HBM to Nvidia, while Micron has also gained ground. Samsung has been working to qualify newer HBM generations and restore its position with the largest AI accelerator customers. A collaboration with TSMC on the logic base die could help narrow a technical gap without requiring Samsung to produce every component internally.
That would also demonstrate a more modular supply-chain model in which memory vendors choose the best available foundry process for specific logic functions. If successful, the approach could be copied in future customized memory products.
Capacity planning will be crucial
TSMC already faces strong demand for advanced logic and packaging capacity from Nvidia, AMD, Apple and other customers. Adding HBM base-die work creates another claim on scarce leading-edge capacity. The commercial impact will therefore depend not only on design performance but on how much 3nm capacity TSMC can allocate and when.
For Nvidia, diversifying HBM supply is strategically valuable because memory shortages can limit GPU shipments even when graphics processors themselves are available. Bringing Samsung more fully into the supply chain could reduce that bottleneck and increase competition among memory vendors.
