NVIDIA CEO Jensen Huang warns memory squeeze will persist for years, reinforcing bullish read-through for Micron and SanDisk

NVIDIA CEO Jensen Huang told investors in South Korea that AI-linked “stocks are very cheap right now” and, more directly for memory shareholders, that the industry’s memory shortage is likely to continue for “several years,” according to a report published Monday. The comments landed after a Friday sell-off that hit semiconductor stocks broadly, with memory-focused names moving sharply in the immediate aftermath.

In the same remarks cited by 24/7 Wall St., Huang said demand across the AI supply chain was “enormous,” adding that shortages extend beyond the chips themselves. He pointed to constraints affecting “wafers to silicon photonics and cable connectors,” framing the problem as supply-side tightness rather than a near-term collapse in AI demand.

For memory investors, the most market-relevant detail was Huang’s multi-year view of the shortage and Nvidia’s stated effort to align next-generation memory with its product roadmap. The report also tied the supply crunch to Nvidia’s upcoming Vera Rubin systems, which Nvidia positions as its platform for agentic AI and long-context reasoning that aims to reduce bottlenecks in communication and memory movement.

Nvidia has previously laid out what that roadmap work looks like in contract and technology terms. In a June 7, 2026 press release, Nvidia announced a multiyear technology partnership with SK hynix to advance next-generation memory for what Nvidia called the global AI factory buildout. The company said advanced memory is essential to AI factories’ performance and that the agreement is designed to address the long development cycles and capital investment needed for advanced memory.

That SK hynix partnership, according to Nvidia, includes codeveloping memory for specific Nvidia platforms, including Vera Rubin AI supercomputers and Vera CPUs, as well as RTX Spark™-powered PCs and Jetson Thor robotic computing platforms. Nvidia also said SK hynix will use Nvidia CUDA-X libraries and PhysicsNeMo to accelerate semiconductor simulations, and that it will develop fab “digital twins” using Nvidia Omniverse, OpenUSD, and cuOpt to support more autonomous semiconductor manufacturing.

The report that carried Huang’s comments also described market turbulence in South Korea and the immediate reaction in U.S.-listed memory names. It said the KOSPI index fell about 4.8% overnight, noting that Korean memory heavyweights like Samsung and SK hynix help dominate the index. It further said Micron and SanDisk were up modestly in overnight trading after the prior week’s sharper decline, with Micron and SanDisk cited as having been among the hardest hit in the pullback.

Still, key specifics were not disclosed in the 24/7 Wall St. write-up. It did not provide quantified supply forecasts, contract volumes, or pricing terms tied directly to Huang’s “several years” comment, and it did not indicate whether Nvidia’s memory guidance would translate into updated purchase schedules for individual memory suppliers.

Investors watching the memory cycle next will likely look for confirmation in earnings calls and customer updates from Micron and SanDisk, as well as any additional details from Nvidia on how its Vera Rubin-related buildout is progressing. The immediate question is whether the shortage Huang described will remain primarily a multi-year capacity constraint or begin to show offsets through new manufacturing ramps and technology transitions. “