TSMC has reportedly begun drafting plans to make 3D DRAM at the request of Nvidia, its largest customer, according to a report by The Information published on 6 November 2026. Neither company had confirmed the plan in published coverage; it is a report based on unnamed sources, not an announcement.
What was reported and when
On 6 November 2026, The Information reported, citing two people with direct knowledge of the matter, that TSMC had begun drawing up plans for advanced memory production. The Next Web, reporting the same day, wrote that TSMC ‘recently began drafting plans to make 3D DRAM’, a design that stacks memory cells vertically to increase capacity.
The reported options include building new factories or buying an existing plant and converting it. The effort is reportedly led by Hsu Kuo-chin, a TSMC vice president who oversees advanced packaging. The coverage discloses no production date, investment commitment or manufacturing location, as eeNews Europe noted on 10 November 2026.
What TSMC and Nvidia have said
The Next Web reported that TSMC and Nvidia did not immediately respond to requests for comment. eeNews Europe reported that TSMC said it does not comment on market rumours or speculation, and that a spokesperson told Tom’s Hardware the company remains focused on its core integrated-circuit foundry business. That statement did not explicitly rule out an internal study. eeNews Europe reported no response from Nvidia and stated that neither company has announced a memory manufacturing programme.
Why a foundry would move toward memory
The Next Web describes memory as in short supply because AI data centres are absorbing it, pushing prices up, and says Micron expects the shortage to last into 2028. A TSMC entry would, in its words, put the company up against Samsung, SK Hynix and Micron. The same article reports that in October 2026 Nvidia reportedly asked Samsung, SK Hynix and Micron to build the base layer of its custom HBM memory on TSMC’s 3nm process, which means TSMC is already moving closer to memory manufacturing.
What it means for research computing and research offices
The following is editorial interpretation, not reported fact. University research computing depends on GPU servers and the memory that feeds them, so a tight memory market can show up in cluster refresh quotes, lead times and the cost of campus GPU nodes. A reported plan with no date or committed investment cannot change that supply picture soon, and the sources do not suggest it would. Research offices budgeting for compute are better served by watching actual supply and pricing signals than by this report alone.
- Allocation models matter more when hardware is scarce: compare national allocations with campus recharge in NSF ACCESS versus campus recharge.
- Large public systems show how much accelerator capacity a national facility concentrates: see the El Capitan supercomputer guide.
- Researchers who need accelerators now can use shared queues: see requesting a GPU in a Slurm batch job.
- Energy and efficiency are part of the same planning conversation: see sustainable HPC.
What is still unknown
The reporting does not say whether TSMC will proceed, when any product could ship, which sites are under consideration, or how Nvidia’s request would be structured commercially. Those points remain open until the companies say something on the record.
Sources
- The Next Web: TSMC is weighing a move into AI memory chips at Nvidia’s request, 6 November 2026, checked 2026-11-15.
- eeNews Europe: TSMC 3D DRAM report linked to Nvidia draws response, 10 November 2026, checked 2026-11-15.







