Skip to navigationSkip to main contentSkip to right columnMicah Zimmerman, The Motley FoolSun, June 28, 2026 at 8:20 PM GMT+2 5 min readAbout two weeks ago, Advanced Micro Devices announced the acquisition of MEXT, a start-up that has built artificial intelligence (AI)-driven software designed to make NAND flash behave like dynamic random-access memory (DRAM).The technology uses predictive algorithms to identify frequently accessed data and move it between flash storage and high-speed memory in real time, reducing the amount of expensive DRAM a data center needs to run AI workloads at scale. According to MEXT's own press release, the software can cut memory costs by nearly half while expanding usable memory capacity by two to four times.Missed Nvidia in 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" signal flashed for a little-known chipmaker called Nvidia. For the first time in years, that same "Total Conviction" signal is flashing for a company 1/100th the size of Nvidia. Continue »For investors in Micron Technology (NASDAQ: MU) and Sandisk (NASDAQ: SNDK), the knee-jerk read is obvious: If AMD can teach flash to behave like DRAM, demand for high-bandwidth memory contracts declines. The knee-jerk read is terribly wrong.What MEXT actually does (and doesn't do)MEXT's technology operates in the software tier between existing storage and compute. It doesn't replace DRAM or HBM. Instead, it reduces the amount of high-speed memory certain workloads require by optimizing what lives in it at any given moment. That's a meaningful efficiency gain for enterprise customers running general-purpose AI workloads, where memory is a cost constraint.What it cannot touch is the physics of training large AI models and running inference at the performance levels that hyperscalers require. An Nvidia Blackwell graphics processing unit (GPU) demands HBM4 not because no one has tried to work around it, but because the bandwidth requirements of training trillion-parameter models are architectural constraints, not software problems. No predictive tiering algorithm changes what the silicon needs.MEXT is a tool for enterprises trying to stretch existing infrastructure. It is not a substitute for the memory products that Micron and Sandisk sell to massive tech companies.Image source: Getty Images.Micron's position is structurally insulatedMicron Technology's entire 2026 HBM4 production is sold out under binding multi-year contracts. At COMPUTEX 2026 in May, the company laid out an end-to-end AI memory portfolio spanning data center to intelligent edge, all in high-volume production. Fiscal first-quarter 2026 revenue hit $13.64 billion, up 57% year over year, with gross margins around 56%, driven by HBM pricing power that comes from contracted scarcity.Terms and Privacy PolicyEU DSA contactPrivacy & Cookie SettingsMore Info