YMTC Plans a Fourth Wuhan Fab for 300-Layer NAND: What Reuters Reported and What Is Confirmed

China's largest NAND flash maker, Yangtze Memory Technologies (YMTC), will reportedly target 60 to 70 percent domestic equipment at a fourth fab in Wuhan. The plant is meant to make NAND chips with more than 300 layers for AI data centers, and construction would start next year. This comes from a Reuters report on November 6, 2026, based on two unnamed sources. We could not open the full Reuters page, so we rely on syndicated copies and on earlier, readable coverage from April. Those copies give the target as at least 60 percent, with an ambition of 70 percent. YMTC has not confirmed the plan in anything we could read.

This is an AI supply-chain story. AI clusters need storage as well as GPUs, and the fight over who can build the memory is tied to export controls.

TL;DR: reported vs confirmed

What did the November 6 report say?

A summary of the Reuters story says YMTC will aim for 60 to 70 percent local sourcing for equipment at its upcoming fourth Wuhan fab. That would be up from a share a little over 50 percent at the third plant. The fourth plant is intended to produce NAND with more than 300 layers for AI data centers. Construction on the fourth site would begin next year, as part of a broader expansion alongside plants three and five. Once all are running, the three newest fabs would each process about 100,000 wafers a month, more than doubling YMTC's capacity despite US export limits.

The Reuters headline, as quoted in that summary, is "China's top flash memory chipmaker targets more homegrown tools for new factory, sources say." The wording matters: "targets" and "sources say" mean this is a plan described by insiders, not a signed commitment.

What did the April reporting add?

Tom's Hardware, reporting on Reuters' April 2026 story, gave more detail that we could read in full:

• Three sources told Reuters that more than 50 percent of Phase 3's tooling came from inside China.
• YMTC aimed to add two more fabs of equivalent scale. Those two were not tied to dates or locations.
• Phase 3 alone would reach 50,000 wafers a month by 2027 and 100,000 at full capacity. Tom's Hardware put the combined capacity of the first two Wuhan fabs at 200,000 wafers a month.
• Tom's Hardware reported that Chinese authorities had begun rejecting approvals for new fabs unless applicants could show, through procurement tenders, that at least half of the equipment would be Chinese-made. The rule is not published in a formal regulation.
• Part of each new fab's capacity is understood to go to DRAM instead of NAND. Tom's Hardware says YMTC decided to allocate 50 percent of Phase 3 capacity to DRAM, citing a supplier quoted by Nikkei Asia.

Reading the two reports together, the November story looks like the follow-up: the "two more fabs" now have a number, 60 to 70 percent local tools, and a product target. That is our reading, not a statement from either source.

Why does local tooling matter?

Tom's Hardware explains why NAND is the easiest memory type for China to localize. 3D NAND scales upward by adding layers rather than by shrinking features. That moves the bottleneck from lithography, where Chinese tools are weakest, to high-aspect-ratio etching, deposition and wafer bonding, where they are stronger. The article names AMEC and Naura as domestic etch suppliers and says Naura supplies etching tools for chips with more than 300 layers.

The weak point remains lithography. Tom's Hardware reports that YMTC still depends on imported lithography tools, and that US lawmakers have introduced the MATCH Act to extend controls to older immersion DUV machines. If that bill moves, YMTC's stockpiled foreign tools become harder to maintain, which strengthens the case for a domestic-tool plan. We have not checked the bill's current status.

How does this connect to AI?

The AI link is the reported product target: NAND for AI data centers. Storage matters in three places: training data pipelines, model checkpoints and retrieval for inference. High-layer NAND gives denser, cheaper flash per rack.

We want to be careful here. The AI framing comes from the summary of sources' description. We found no YMTC datasheet, customer name or capacity commitment for AI storage. The reported capacity numbers also describe wafers, not finished drives.

The wider context is a crowded week for AI hardware supply stories. See our coverage of Samsung's second Taylor fab and HBM base dies, Micron pausing its Taoyuan plant, and China's 24 GW compute capacity versus 56 GW in the US. On the GPU side, Nvidia's H200 stockpile in Hong Kong shows Beijing's own import choices, and Tencent's Oracle chip lease shows how buyers work around export rules.

Timeline: how the Wuhan expansion reached this point

Two points stand out. First, YMTC's last confirmed leading product in these sources is a 232-layer part from 2022, so a 300-plus-layer target is a step forward on paper, and there is no independent test of it. Second, the domestic-tool bar rose from 50 percent as a baseline to a reported 60 to 70 percent target within seven months, which fits the direction Beijing is described as setting.

What does "300 layers" actually mean?

3D NAND stacks memory cells vertically. More layers means more bits in the same footprint, and each added layer makes the etching step harder, because the machine must cut clean channel holes through a very tall stack. Tom's Hardware describes stacks of 7 to 10 microns of dielectric at 300-plus layers.

That is why layer count is a useful, if blunt, measure of manufacturing skill. It is not the same as performance in a data center. Drive speed, endurance and price depend on the controller, the interface and yields too. A layer count tells you what the fab can build, not what you can buy.

What this means for builders

Most readers do not buy NAND wafers. Three practical points follow.

1. Storage pricing. If YMTC doubles capacity, global NAND supply grows later in the decade. Price effects are uncertain and depend on yields, export rules and demand. Do not plan budgets around this report.
2. Supply-chain risk. If you sell hardware to or from China, a YMTC-based component sits on a US Entity List name. Check your compliance team's guidance, because rules differ by product and customer.
3. Signals, not facts. Reported fab plans often slip. The third fab's start date has been reported variously as 2027 (Nikkei, via Tom's Hardware in November 2025) and late 2026 (Reuters, in April). Treat the fourth fab's schedule with the same caution.

How to check the report

1. Read the Reuters article from your own access and note which lines are attributed to sources.
2. Look for a YMTC filing or statement. YMTC did not respond to Reuters' request for comment on the April plan, per Tom's Hardware's account.
3. Watch for local government approvals in Wuhan and for tender notices from Chinese tool makers, which would show orders.
4. Check whether the 300-plus-layer product shows up on a YMTC product page or at an industry event.
5. Compare the 60 to 70 percent target against any later report of the third fab's actual domestic share.

What is still open

• Whether the 60 to 70 percent tool target is realistic without domestic lithography.
• Whether the fourth fab's start date holds, given the conditional language reported in April about follow-on fabs.
• How much of the new capacity goes to NAND versus DRAM and HBM-related packaging.
• Whether new US controls change the plan.
• Whether YMTC publishes product specifications for the AI-focused chips.

Related reading

• Samsung's second Taylor fab and HBM base dies
• Micron pauses Taoyuan plant
• China 24 GW compute vs US 56 GW
• Nvidia H200 stockpile in Hong Kong
• Tencent leases 100,000 AI chips from Oracle
• Nvidia chip tracking for Southeast Asian partners

Details are accurate as of November 6, 2026. This story rests on anonymous sourcing and may change.