NEW YORK, March 4, 2026, 10:06 (EST)
- Micron has started sending out customer samples of its 256GB low-power server memory module, targeting AI data centers.
- The company touted reduced power consumption and a smaller size compared to typical server memory, saying some AI workloads also saw quicker response times.
- Micron shares bounced Wednesday, clawing back some ground after dropping roughly 8% in Tuesday’s wider market slide.
Micron Technology, Inc. (MU.O) climbed roughly 3% out of the gate on Wednesday. The stock move comes after the memory-chip company revealed it’s begun shipping customer samples of a new 256GB low-power server memory module, targeting AI data centers.
Timing is crucial right now, as data center operators run up against real constraints on power and cooling. Memory, which serves as the short-term storage feeding processors, is turning into a bottleneck for AI systems. Pricey chips sometimes just wait, idling, if data doesn’t show up in time.
Micron is rolling out a module built with LPDRAM—low-power dynamic random access memory—built for reduced energy consumption. The company’s latest comes in a removable server module dubbed SOCAMM2, a modern form factor aimed at packing more memory into tighter quarters.
Micron is touting its 256GB SOCAMM2 as the first in the industry to use a monolithic 32Gb LPDDR5X setup, supporting as much as 2TB of LPDRAM for each eight-channel server CPU. The company’s own tests pointed to quicker “time to first token” for long-context inference jobs—essentially, large language models start spitting out answers faster. “The first to deliver a 32Gb monolithic LPDRAM die,” said Raj Narasimhan, Micron’s senior vice president. Nvidia’s Ian Finder, for his part, made the case that AI infrastructure requires “optimization at every layer.” 1
Micron shares climbed 2.7%, recently trading at $390.10 and putting the chipmaker’s market value near $258 billion.
Tuesday’s session saw the stock tumble roughly 8% as U.S. equities got swept up in a global rout, with Micron landing on the list of prominent laggards. 2
Samsung Electronics and SK hynix are both moving quickly toward SOCAMM2-type modules, with the sector looking to boost memory in each server while keeping power use in check, according to TrendForce. The firm pointed out that Micron is one of the companies developing solutions that skirt the new JEDEC standard, which governs many memory specs. 3
SOCAMM2 has been getting noticed lately, with buyers of AI servers zeroing in on its answer to persistent complaints about power consumption and heat—issues that go well beyond the chase for pure compute muscle. According to Tom’s Hardware, the form factor isn’t just a random choice; it’s synced up with Nvidia’s server roadmap. The pitch: squeeze more memory per CPU and cut down on energy use compared to the standard RDIMMs—the server “sticks” everyone is running in racks today. 4
Micron will release its fiscal Q2 numbers March 18, with the earnings call set for 2:30 p.m. Mountain time, according to the company. 5
Still, adoption is the sticking point. Micron backs its speed and power claims with internal tests, but everything rests on how fast server vendors and cloud buyers shift to SOCAMM2 designs. Rivals could also squeeze on pricing if the memory cycle flips.
Investors now have their eyes on SOCAMM2—will it jump from sampling to volume orders quickly enough to make a dent in revenue? Micron’s margin defense gets put to the test, too, as Samsung and SK hynix target the same AI server sockets.
