The race for artificial intelligence isn’t just about GPUs. Increasingly, the bottleneck is less glamorous but equally decisive: memory. While part of the industry focuses resources on high-margin data center products (like HBM in DRAM) and enterprise demand ramps up again, a strategic rumor is circulating in Asia: SanDisk and Micron are exploring agreements to leverage available capacity at PSMC’s new 12-inch factory in Tongluo, Taiwan.
The idea itself is revealing. It’s uncommon for memory giants—with their own manufacturing cultures—to look externally to third parties to quickly open capacity. But the current context is forcing less orthodox decisions: expanding production organically is slow, expensive, and constrained by machinery, specialized labor, and permits; obtaining ready-made capacity can be the fast track, even if it involves complex cooperation models.
A key asset: PSMC’s 12-inch plant in Tongluo
PSMC launched its 12-inch wafer plant in Tongluo (Miaoli County) as a large-scale project: over NT$300 billion in investment, with construction starting in March 2021 and an initial phase focused on trial production.
According to the company, the facility is designed to operate on mature nodes (55 nm, 40 nm, and 28 nm), with an anticipated capacity of about 50,000 12-inch wafers per month in this initial setup. These figures are significant for two reasons:
- Mature nodes remain critical for a wide range of components (controllers, peripheral circuitry, automotive, industrial, consumer electronics), and also for parts of the storage/memory ecosystem where not everything needs the “latest nanometer”.
- In a market where top-tier factory capacity is highly contested, large volumes at 28–55 nm can be gold if they enable rapid scaling of products that don’t depend on the most advanced nodes.
What the rumors say: “quick capacity” and various options under discussion
Taiwanese financial media have reported that SanDisk has contacted PSMC and that, subsequently, Micron has also initiated discussions to explore cooperation. The focus is on Tongluo’s plant, described as an asset already built but with room to add capacity relatively quickly.
The published information suggests:
- Micron is studying several approaches (at least three options), one being introducing technology and equipment to produce memory and activate capacity “as quickly as possible”.
- PSMC, meanwhile, neither confirms a finalized agreement nor denies ongoing conversations with multiple companies about collaborations.
- In parallel, there are reports that the discussion might go beyond simple manufacturing contracts, including options like structured collaborations or even corporate operations (e.g., acquisition or equity stake), though these are market rumors rather than confirmed facts.
An important nuance: these leaks coincide with a very “hot” investment climate around the memory sector in Taiwan, attributed to strong demand linked to AI and capacity shifts within DRAM (some industry players are shifting lines toward AI products), fueling a narrative of supply tension.
Why would this be “unusual” yet logical?
It’s not common because historically, major memory manufacturers prefer to control their fate: own factories, own processes, own schedules. Outsourcing manufacturing—or relying on third parties to rapidly open capacity—can introduce dependence, integration friction, and reputational risk if yield or quality problems occur.
But it can make sense today for three very practical reasons:
- Speed: building a new factory isn’t a project of months; it takes years. Leveraging existing capacity shortens the cycle.
- CapEx and financial discipline: even for industry giants, committing billions to new manufacturing lines is risky in cyclical markets.
- Geopolitical flexibility: diversifying industrial footprint and supply options has become a survival strategy, not just a matter of efficiency.
The SanDisk context: a “retooling” company after spinning off from Western Digital
In SanDisk’s case, there’s an additional corporate element that helps interpret this move: the brand and business operate as a separate company following the spin-off from Western Digital, with its own listing and market narrative. In this scenario, securing capacity and ensuring a product roadmap can be part of a “new phase” strategy, especially if the market rewards those who can deliver volume reliably.
What PSMC gains and what risks it faces
For PSMC, an agreement with players like SanDisk or Micron would entail:
- Better utilization of a huge asset and potentially long-term contracts.
- Positioning: moving from “mature node capacity” to becoming a tactical part of the memory supply chain.
The risk is the opposite: if collaboration requires process adjustments, governance changes, or prioritization of one client over others, it could strain their portfolio. And if the deal involves investment or equipment supplied by the partner, careful consideration is needed around who controls decision-making, who bears industrial risk, and how the returns are shared.
What to watch for in 2026
PSMC emphasizes it will continue on its existing roadmap, with plans for silicon interposers and 3D ICs related to their expansion, even as talks about potential agreements proceed. In practice, three signals could indicate the rumor is turning into reality:
- Official announcement (joint venture, manufacturing agreement, capacity purchase, or acquisition).
- Evidence of equipment installation and measurable ramp-up at the plant.
- Indirect confirmation: production guidance, CapEx plans, and public forecasts from partners.
The recent AI surge has shown that “invisible infrastructure”—memory, interconnection, packaging—is as determinant of winners as computing silicon. Movements like Tongluo, even if born as rumors, are being closely monitored.
Frequently Asked Questions
What is a “12-inch” factory and why does it matter for memory and semiconductors?
It refers to the wafer diameter (300 mm). Larger wafers enable more chips per cycle, improving economies of scale and industrial capacity.
Is it sensible to produce memory-related products on nodes like 28 nm, 40 nm, or 55 nm?
Yes, depending on the component. Not everything is “leading-edge.” Many peripheral circuits, controllers, and parts related to storage and industrial electronics are manufactured on mature nodes due to cost, availability, and reliability.
What real options exist if a company wants to “buy” capacity from another?
From manufacturing agreements (foundry/wafer supply) and capacity reservations to joint ventures, line-sharing with equipment contributions, or even corporate acquisitions, depending on the case.
Can such an agreement help alleviate supply pressure from AI data centers?
It can support specific segments and products reliant on that capacity. But the overall impact depends on what’s actually manufactured, the ramp-up pace, and whether the main market bottleneck is in DRAM/HBM, NAND, or other parts of the supply chain.