Solidigm weighs the first US NAND fab as the memory squeeze spreads

Share
Solidigm weighs the first US NAND fab as the memory squeeze spreads

The AI memory shortage has been a DRAM and HBM story all year. Storage is next, and both Washington and Beijing are moving on it in the same week.

SK Hynix's US subsidiary Solidigm is weighing what would be the first NAND flash factory on American soil, with upstate New York emerging as a leading candidate site. The company is studying potential locations, according to reports citing people familiar with the plan, and no decision has been announced. The move would put NAND — the flash that holds model weights, checkpoints, vector stores and the training data lakes behind them — in the same onshoring push that has already pulled advanced packaging and HBM assembly to the US.

Why it matters: NAND spent years as the semiconductor industry's loss-making laggard, and it is now the cheapest part of the AI storage stack to run short of. Enterprise SSD demand from data centers has flipped the market, and the same squeeze that made DRAM and HBM the tightest links in the supply chain is now reaching flash. SK Hynix is also separately in talks with Intel about manufacturing memory in Ohio, and it has already committed to building out capacity in Korea. A US NAND plant would give Solidigm, the ex-Intel flash business SK Hynix bought outright, a domestic footprint at the moment its parent is deciding how much of the AI buildout it wants to serve from inside the US.

We covered the DRAM and HBM side of this squeeze two days ago — SK Hynix and Intel discuss the first US-made memory chips — and the layer has moved again: this time it is flash, not HBM.


China's answer arrived the same day. ChangXin Memory Technologies, the country's leading DRAM maker, is preparing to move into NAND flash, directly into the market of domestic rival YMTC and of Samsung and SK Hynix. Reuters reported that CXMT is setting up a NAND research line in Beijing, a first step toward a full product line. The timing is not accidental: memory prices have risen far enough that a new entrant can justify the capital, and Beijing has made storage self-sufficiency a standing priority.

The strategic read is a familiar one. Every layer of the AI stack that gets scarce gets localized — by the country that has it and by the country that doesn't. CXMT entering NAND makes China's memory industry vertically complete in a way it wasn't a year ago, and it puts price pressure on the incumbents precisely when they are spending record sums on capacity. Watch whether CXMT's line stays a research effort or converts to volume: that is the difference between a headline and a supply shock.

What to watch: whether Solidigm confirms a site and a timeline, and whether the US treats a NAND fab as strategically urgent or merely commercially welcome.

If every scarce layer of the AI stack ends up built behind a national border, what happens to the price of training a model? Write to us via our contact page.

Get SiliconNoon in your inbox

One email a day with the AI stories that matter — what happened, the numbers, and why it matters. No spam, unsubscribe in one click.

Read more

Nvidia's B200 now resells for 58% above list — and lenders noticed

Nvidia's B200 now resells for 58% above list — and lenders noticed

A year-old AI accelerator is appreciating like real estate, not like a graphics card. That changes what gets borrowed against it — and what it says about the shape of the boom. Nvidia's B200 is trading at 158% of its launch price roughly a year after it reached mass availability, according to Silicon Data's analysis — an accelerator that added 58% instead of shedding the 20% to 40% that standard depreciation schedules assume.

Huawei names its 2027 Nvidia challengers: the 960DT and Ascend 960PR

Huawei names its 2027 Nvidia challengers: the 960DT and Ascend 960PR

Huawei put dates and part numbers on its AI chip roadmap for the first time on Thursday. In the same hour, two labs published work that shrinks the machine a 35B model needs and shows what coding agents can build when nobody hands them a strategy. Huawei's rotating chairman David Wang told a company event in Shanghai that Huawei will ship the 960DT in the first quarter of 2027 and the Ascend 960PR in the third quarter, naming two specific parts for the first time rather than describing a…

Alibaba plans a 10T-parameter model and a 500,000-chip cluster

Alibaba plans a 10T-parameter model and a 500,000-chip cluster

Alibaba's cloud conference in Hangzhou opened this morning with a three-part commitment that is harder to wave off than the usual keynote arithmetic: a model bigger than anything trained so far, the chip to run it, and the power to feed both. Huawei spent the same week putting a coding model inside its own operating system, and a small project on the local-AI forums quietly showed that a model's refusals can be switched off without touching a single weight. Alibaba says its next Qwen models…

Alibaba opens cloud regions in Turkey and Finland as AI tensions bite

Alibaba opens cloud regions in Turkey and Finland as AI tensions bite

Alibaba picked the week of a US–UK AI pact to plant its cloud deeper in Europe. The same conference where it announced the regions also carried its chip roadmap — and Japan's turbine makers are still selling the power behind all of it. Alibaba Cloud will open its first cloud regions in Turkey and Finland — plus one in the Netherlands — within the next 12 months, the company said at its Apsara conference in Hangzhou on Wednesday.

China Mobile Cloud runs attention on GPUs and FFN on brain chips

China Mobile Cloud runs attention on GPUs and FFN on brain chips

Two announcements out of China this week point the same direction: when the chip node is capped, the wins have to come from system architecture — and from a build-out policy that finally measures utilization instead of volume. China's first production system that splits a large model across domestic GPUs and brain-inspired chips went live today, and the vendor claims it more than doubles inference output while cutting operating cost by more than 40%.