Micron, Samsung and SK Hynix hit $1T on AI memory. UC Merced feels it.

Key Takeaways

Micron, Samsung and SK Hynix now sit near or above $1 trillion in market value.
High-bandwidth memory demand for AI systems pushed those gains, the Times reported.
UC Merced’s H200 GPU nodes rely on HBM3, tying campus compute to that same supply chain.
California hosts about 5% of U.S. data center capacity, so expansion pressure moves inland.

141 gigabytes. That’s the high‑bandwidth memory on each of UC Merced’s new H200 GPUs, according to campus IT docs. If you’re wondering why that number suddenly matters in Merced County, the New York Times just laid out the answer: memory chips are the new center of gravity in the AI boom, with Micron, Samsung and SK Hynix hitting the trillion-dollar tier this spring.

The Valley angle is simple. Those HBM parts feed the same GPUs researchers, startups and agencies here try to rent or buy. When memory is tight, cycles get harder to book and more expensive.

What the Times reported

DealBook’s rundown on May 27 said Micron and SK Hynix joined the $1 trillion club as investors piled into suppliers of high‑bandwidth memory, the stacked chips that sit next to Nvidia’s data‑center processors. Samsung also reached the mark this month, putting all three memory giants in rarified territory. Washington is paying attention too. The U.S. trade representative signaled new tariff talk aimed at shoring up domestic chipmaking, and Samsung’s labor dispute ended with profit‑sharing tied to AI chip results. The article framed a market that has pivoted from GPU logos to the memory behind them.

Why this lands in the Valley

UC Merced’s Pinnacles cluster lists nodes with NVIDIA H200 NVL cards, each carrying HBM3 memory. That’s the same component class in short supply and under investment glare. When HBM capacity loosens or tightens, the ripple hits cloud pricing and campus queues. Fresno State is pushing an AI campus effort, and local ag‑tech teams often rent compute by the hour. They feel the same ripple.

California has a small slice of the country’s data‑center footprint, roughly 5% by a recent analysis, so big new server farms often land in cheaper, power‑available regions. That includes interior corridors like ours when the fiber is there. If memory makers keep pace and GPU vendors ship on time, pressure shifts from chips to siting, power and cooling. Then the questions move to the council and the supervisors.

A floor fan hummed by the server room door.

What to watch next

Supply: Micron says it’s investing heavily in HBM capacity. If those lines run as planned, campuses and midsize firms should see better availability over the next few quarters.
Policy: Any new tariff on imported chips would change price curves. That could help Boise‑based Micron on paper, but it can raise near‑term costs for integrators and buyers downstream.
Labor and logistics: Samsung’s strike talk ended, but packaging, testing and shipping remain tight. One snag in Korea or Taiwan can freeze deliveries here.
Research pivots: UC Merced and partner labs keep publishing on ways to cut memory pressure in training and inference. If those methods catch on, the local need tilts from the biggest GPU to smarter scheduling.

The local bottom line

If you’re a Valley researcher, IT lead or startup accountant, watch memory, not just GPUs. It’s the part of the stack that’s deciding who gets to run big jobs and when. The lights on the Pinnacles rack tell the story either way, green and steady in the cold.

Central Valley AI is produced by the CVAI Newsdesk team and developed by Kaweah Tech, a regional firm that builds, deploys, and integrates AI solutions for businesses across California's Central Valley.

Source

https://www.nytimes.com/2026/05/27/business/dealbook/ai-chips-war-samsung-micron.html