Taiwan Semiconductor Manufacturing Company (TSMC), the world’s largest contract chipmaker, is reportedly exploring a move into advanced memory manufacturing after a request from Nvidia, its biggest customer. The company has begun drawing up plans to produce a next-generation memory technology called three-dimensional dynamic random-access memory (3D DRAM), according to a November 6 report by The Information. The potential expansion could take TSMC beyond its traditional foundry business and into a semiconductor market dominated by Samsung Electronics, SK Hynix and Micron Technology.
The reported move comes as the artificial intelligence boom drives unprecedented demand for processors and memory used in data centres. Shortages of advanced memory components have pushed up prices and prompted major technology companies to look for additional sources of supply. If TSMC eventually moves into memory production, it could strengthen its position across more stages of the AI hardware supply chain. However, the company has not announced a confirmed investment or production schedule, and its official response reiterated its focus on its core foundry business.
Why TSMC Is Considering Entering the Memory Business
According to The Information, TSMC has recently started developing plans to manufacture 3D DRAM at Nvidia’s request. The report cited two people with direct knowledge of the matter.
The initiative is reportedly being led by Hsu Kuo-chin, a TSMC vice president who oversees advanced packaging operations. The company is examining different ways to establish manufacturing capacity, including building new factories or acquiring an existing semiconductor facility and converting it for memory production.
No final decision has been reported on the scale of investment, the manufacturing location or when production could begin.
The potential move reflects a growing challenge for AI chipmakers: securing enough memory to support increasingly powerful processors. Modern AI accelerators require enormous quantities of fast memory to process information efficiently. As demand for AI infrastructure expands, memory manufacturers have faced pressure to increase production capacity.
Nvidia relies on external suppliers for memory used alongside its GPUs. A potential TSMC entry could create another source of supply for advanced memory technology and give Nvidia greater flexibility in its component ecosystem.
What Is 3D DRAM and How Does It Work?
DRAM, or dynamic random-access memory, is a type of semiconductor memory that temporarily stores data while a processor uses it. It is widely used in computers, servers, smartphones and other electronic devices.
Traditional DRAM stores memory cells in a largely horizontal arrangement. As demand for higher memory capacity increases, manufacturers are exploring ways to use vertical structures to fit more memory into a limited physical area.
3D DRAM aims to stack memory cells vertically rather than relying primarily on a flat structure. This could increase memory density and potentially improve the amount of data that can be stored within a given footprint.
The technology remains under development, and its commercial advantages will depend on manufacturing progress, power consumption, performance and production costs.
It is important not to confuse 3D DRAM with high-bandwidth memory (HBM). Both involve vertical stacking concepts, but they are different technologies.
| Technology | Structure | Primary purpose |
|---|---|---|
| Conventional DRAM | Memory cells arranged largely in a planar structure | General-purpose memory |
| HBM | Multiple DRAM dies stacked and connected vertically | High memory bandwidth for AI and high-performance computing |
| 3D DRAM | Memory cells themselves arranged in vertically integrated layers | Higher memory density |
HBM is already widely used in AI accelerators, while 3D DRAM is a next-generation architecture that still faces significant development and manufacturing challenges.
Nvidia’s Request Reflects Growing Pressure on AI Memory Supply
Nvidia’s reported request highlights the importance of memory availability in the AI infrastructure market.
The company designs GPUs that power AI training and inference systems. These processors require high-performance memory to move large volumes of data quickly. Even when GPU manufacturing capacity expands, shortages of memory can limit how many complete AI systems can be delivered.
The memory supply chain is currently dominated by Samsung Electronics, SK Hynix and Micron. These companies are investing in advanced memory technologies, including HBM and next-generation DRAM architectures.
Adding TSMC as a potential memory manufacturer could provide another participant in the market. It may also give Nvidia more options as it tries to secure the components required for future AI systems.
The move would not necessarily replace existing memory suppliers. Instead, TSMC could eventually become an additional source of advanced memory, depending on whether its technology proves commercially viable.
TSMC Could Challenge Samsung, SK Hynix and Micron
A move into memory manufacturing would represent a major strategic development for TSMC.
The company has built its business around manufacturing chips designed by other companies. Its foundry services are used by major technology businesses, including Nvidia, Apple and AMD.
Memory manufacturing, however, operates under a different competitive structure. Samsung, SK Hynix and Micron have established production facilities, specialised technologies and long-standing customer relationships in DRAM.
If TSMC enters this market, it would have to compete with companies that already possess extensive experience in memory manufacturing.
The potential expansion could also change the relationship between foundry and memory suppliers. TSMC currently plays an important role in manufacturing logic chips and advanced packaging, while other companies supply much of the memory integrated into AI hardware.
Producing its own memory could allow TSMC to participate in more stages of the manufacturing process. However, it could also create new competitive tensions with companies that are currently its partners and customers.
Manufacturing Challenges Could Delay Commercial Production
Despite the strategic opportunity, producing 3D DRAM at commercial scale will not be straightforward.
Manufacturing vertically integrated memory cells requires precise control over materials, electrical connections and individual layers. As the number of layers increases, manufacturers must maintain consistent performance and minimise defects throughout the structure.
Production yields will be particularly important. Even technically successful memory designs may not become commercially attractive unless manufacturers can produce them reliably and at competitive costs.
TSMC would also need to decide whether to build new factories or convert existing facilities. Constructing new semiconductor plants requires substantial capital investment and can take years before production begins.
Acquiring an existing facility could provide another route, but converting a plant for a different type of semiconductor manufacturing also involves technical and financial challenges.
The memory industry is additionally known for its cyclical nature. Periods of strong demand and high prices can encourage manufacturers to expand capacity, but new supply may eventually contribute to oversupply if demand slows.
These factors make it difficult to determine whether TSMC’s reported plans will lead to a large-scale commercial business.
TSMC Reiterates Its Focus on Foundry Operations
Following the report, TSMC reiterated that its focus remains on its core business.
In a statement reported by The Information, the company said it does not comment on market rumours or speculation. It also emphasised its commitment to being a technology and capacity provider for the global logic integrated circuit industry.
Nvidia did not respond to requests for comment, according to the reporting.
The response means the reported memory initiative should be treated as an exploratory plan rather than a confirmed business expansion. There has been no official announcement of a factory, investment amount, production capacity or commercial launch date.
The Bigger Picture
The reported development highlights how the AI boom is changing the semiconductor industry’s competitive boundaries. Memory, advanced packaging and processor manufacturing are increasingly interconnected, as AI systems require faster data movement and more sophisticated hardware integration.
TSMC already occupies a central position in the AI chip supply chain through its foundry and advanced packaging capabilities. Moving into 3D DRAM could extend that influence, but it would also expose the company to the investment requirements and cyclical risks of the memory market.
For Nvidia, the potential involvement of TSMC could offer another route to securing advanced memory as AI infrastructure demand grows. For Samsung, SK Hynix and Micron, it could eventually introduce a powerful new competitor in a market where manufacturing expertise and production yields are critical.
Looking Ahead
The next important developments will be whether TSMC moves beyond preliminary planning, selects a manufacturing strategy and commits capital to 3D DRAM production. The technology must also demonstrate that it can be manufactured at competitive costs and deliver meaningful advantages over existing memory architectures.
For now, TSMC’s potential entry into the RAM business remains an early-stage possibility rather than a confirmed expansion. Nevertheless, Nvidia’s reported request and the wider memory shortage show how urgently AI companies are seeking additional sources of advanced semiconductor components. If TSMC succeeds, the move could reshape competition across the global AI hardware supply chain over the longer term.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
