TSMC Mulls 3D DRAM Entry, and Memory Stocks Blink First

Taiwan Semiconductor Manufacturing Company is considering a push into advanced memory chips, with internal plans focused on 3D DRAM, according to a report from The Information. The report, published November 6, 2026, frames the effort as being pursued at the request of Nvidia, TSMC's largest customer. No launch, no fab, no dollar figure — just plans on paper.

That was enough to rattle the memory aisle. Shares of Micron (MU) and SK Hynix fell after the report surfaced, per The Information's findings. Investors didn't wait for confirmation or a timeline. They moved on the possibility of a new high-volume supplier arriving in a market that has been running hot.

The backdrop is a genuine supply crunch. DRAM and high-bandwidth memory, the stacked chips that sit beside AI accelerators and feed them data, have been squeezed by AI demand, pushing prices higher. Memory now ranks alongside logic chips and advanced packaging as one of the chokepoints slowing the wider semiconductor industry. Apple flagged rising memory costs earlier in 2026, a reminder the squeeze reaches consumer hardware too.

For TSMC, this would be a departure. The company made its name as a logic foundry, building processors designed by others rather than selling its own branded chips. Memory has long been a separate trade run by specialists. A 3D DRAM move would nudge TSMC toward a more integrated model, where more of the AI hardware stack gets made inside its own walls.

The report sketches a second road as well: partnerships with existing memory makers, letting TSMC expand its footprint without standing up an entire memory operation from scratch. Either way, the strategic ripples could redraw alliances across the memory space and pull fresh investment into technologies built for AI workloads.

The stakes split by seat. For memory investors, the near-term story is volatility as the market prices in added competition and potentially softer prices from higher supply. For AI chip buyers, Nvidia included, more memory could loosen a constraint capping how fast hardware ships. Watch for official word from TSMC, any disclosed memory partnership, and any comment from Nvidia on where it sources memory.