The Information reported on November 6 that TSMC has begun considering a plan to manufacture next-generation "3D DRAM" memory at the request of its largest customer, Nvidia. TSMC issued a statement on the 7th (local time), and several outlets reported that it had "denied" plans to enter the market. But the statement actually reaffirmed the company's focus on its core business as a pure-play foundry, and it does not clearly deny a 3D DRAM manufacturing plan. Meanwhile, even if TSMC were pursuing such a plan, some observers estimate it would take at least five years to help ease the memory shortage. However the statement is read, the report is unlikely to improve the memory supply shortage anytime soon.
Did TSMC's statement deny entering 3D DRAM?
Citing two people familiar with the matter, The Information reported that TSMC had started drawing up plans to manufacture 3D DRAM at Nvidia's request. 3D DRAM is a next-generation DRAM technology that raises storage density by stacking memory cells vertically. According to South Korea's Kyunghyang Shinmun, the plan is being led by a vice president who heads TSMC's advanced packaging division, and options reportedly under consideration include building a dedicated new memory fab or acquiring an existing fab and converting it. The English edition of the Chosun Ilbo also reported that Nvidia, facing memory shortages and rising prices, asked TSMC to manufacture next-generation 3D DRAM.
In response, TSMC released a statement explaining its business policy. According to the statement as quoted by BigGo Finance, TSMC said it "has consistently focused on its core business of pure-play IC manufacturing," and that "our mission is to continue providing reliable technology and production capacity to the global logic IC industry." However, neither of the two quoted sentences mentions DRAM or Nvidia. At least from the wording of the statement, it is impossible to tell whether the reported manufacturing plan exists.
Media interpretations differ on this point. Kyunghyang Shinmun noted that TSMC gave no specifics on whether it had received a request from Nvidia or whether it would enter the memory business, including DRAM. BigGo Finance likewise pointed out that the statement does not even contain the word "DRAM." The English edition of the Chosun Ilbo, by contrast, reported the same statement as TSMC denying plans to enter the DRAM market. That reading goes further than what can be taken directly from the text.
What can be said with certainty is limited to this: TSMC reconfirmed its stance of treating contract manufacturing of logic semiconductors as its core business. The statement says nothing about whether it received a request from Nvidia, whether it is weighing a manufacturing plan internally, or whether it has abandoned such consideration. The statement therefore cannot be used to conclude that no plan to enter 3D DRAM exists.
Even if TSMC makes 3D DRAM, relief for the shortage is years away
Even if TSMC is advancing its consideration of 3D DRAM production, it is unlikely to quickly improve the current memory supply shortage. According to an analysis from Tom's Hardwarecited by MLQ, the three major memory makers, Micron, Samsung, and SK hynix, are all developing 3D DRAM, but full-scale commercialization is still some way off. For example, Micron's technology roadmap still includes at least four more generations of conventional planar DRAM. In other words, there is still room to improve existing planar DRAM, and the shift to 3D DRAM will not begin immediately.
TSMC also faces the problem that it cannot build on an existing DRAM mass-production base. Although it has advanced manufacturing technology as the world's largest semiconductor foundry, it does not own planar DRAM mass-production processes. Because the memory cell structure and manufacturing processes differ between logic chips and DRAM, leading-edge logic manufacturing capability cannot simply be repurposed for DRAM. Taking these circumstances into account, Tom's Hardware assesses that even if TSMC pursued a 3D DRAM manufacturing strategy, it would take at least five years to resolve the current memory shortage. That timeframe, however, is not a development plan announced by TSMC but a media estimate that takes technical challenges into account.
Kyunghyang Shinmun also points out that 3D DRAM still faces challenges in both technology development and mass production. Beyond establishing memory cell stacking technology, a company would need to manage yield and manufacturing cost at volume and build a production system capable of competing with established DRAM makers. TSMC's technological edge in semiconductor manufacturing does not guarantee early mass production of 3D DRAM.
That said, TSMC is not doing zero research on memory-related technology. According to BigGo Finance, at the 2025 VLSI Technology Symposium, TSMC presented technology for embedding "1T1C" memory cells, consisting of one transistor and one capacitor, in the interconnect layers of a logic chip. A typical SRAM cell requires six transistors to store one bit, whereas the 1T1C structure is better suited to shrinking the storage element and could raise on-chip memory density.
However, this is technology for embedding memory in logic semiconductors, which is different in nature from a business that mass-produces standalone DRAM products. TSMC has not disclosed when the technology might be commercialized, nor has it said whether it would position it as a DRAM product. This research record therefore cannot be treated as evidence supporting the reported 3D DRAM manufacturing plan.
For PC makers, smartphone makers, and memory buyers in Japan, what matters is when actual supply and prices will improve. Following Tom's Hardware's analysis, even if TSMC newly entered the 3D DRAM market, the effect would not appear in the market for several years. On the basis of this report alone, it is hard to expect a near-term increase in DRAM supply or a fall in prices.
The view that TSMC denied entry, and what to watch next
There are grounds, though, for interpreting TSMC's statement as an effective denial of entry. The English edition of the Chosun Ilbo reported the statement as a denial of plans to enter the DRAM market, and Tom's Hardware offered a similar interpretation. According to BigGo Finance, Chao Yu-chun, chairman of Taiwanese memory maker Winbond Electronics, said that direct entry by TSMC seems unlikely because DRAM manufacturing involves proprietary process technology and a complex web of patents.
Business interests are also important in judging the likelihood of entry. The English edition of the Chosun Ilbo points to TSMC's long-standing policy of not competing with its customers. If TSMC entered DRAM manufacturing in earnest, it could end up competing with Samsung, SK hynix, and Micron, which are also its current clients. Kyunghyang Shinmun, also citing Tom's Hardware's analysis, mentions that entering the memory market could affect TSMC's collaborative relationships with companies such as SK hynix. For TSMC, the risk of changing its relationships with existing customers must be weighed alongside the profits from a new business.
However, this does not rule out the possibility that TSMC could manufacture some memory-related products while collaborating with existing memory makers. In fact, in October, South Korean media reported that Samsung would use TSMC's 3nm process to manufacture the base die for its next-generation custom HBM for Nvidia (MLQ). The HBM base die is the chip that handles connection to and control of the stacked memory dies, and it differs from the memory die itself, which contains the DRAM storage cells. A memory maker outsourcing base-die production to TSMC needs to be distinguished from TSMC independently entering the DRAM manufacturing business.
Two developments are worth watching in evaluating this report. First, whether TSMC sets out a clear policy on entering DRAM or memory manufacturing in an official announcement or business briefing. Second, whether a contract emerges for TSMC to manufacture DRAM or 3D DRAM memory dies for a specific customer. If either is confirmed, the interpretation of this statement, and of how far TSMC will get involved in the memory business, would need to be revisited.
At this point, there is no confirmed fact showing that TSMC either formally acknowledged or clearly denied a 3D DRAM manufacturing plan. At the same time, even if such a plan exists, establishing mass-production technology and expanding production capacity would take considerable time. What matters in this report is to separate the possibility that TSMC's business scope expands in the future from the timing of any resolution to the current memory shortage. Going forward, the focus will be on whether information emerges that backs up not just whether TSMC is entering the market, but also the specific products it would make, its customer contracts, and its mass-production timeline.