Advanced Micro Devices said Monday it has bought MEXT, a startup whose software helps computers squeeze far more usable memory out of cheaper storage chips, as the chipmaker races to relieve one of the biggest bottlenecks in the artificial intelligence boom. In its announcement, AMD said memory has become “a critical constraint across cloud and enterprise environments,” and that MEXT’s technology will be built into its data center products. The company did not say how much it paid.
The timing was good for AMD shareholders. The stock jumped more than 6% on Monday to about $547.81, and the company’s market value pushed past $900 billion for the first time, helped along by a separate new product aimed squarely at rival Nvidia. But the MEXT deal speaks to a quieter problem that is starting to define the AI era: there is not enough fast memory to go around.
Here is the issue in plain terms. AI systems need two kinds of chips to think. One does the calculating. The other — memory — holds the data those calculations run on. The fastest memory, called DRAM, is expensive and, right now, in painfully short supply, with prices climbing as every company building AI data centers fights for the same chips. A cheaper, more plentiful kind of storage, called flash, is much slower.
What MEXT has built is a workaround. Its software uses AI to predict what data a system will need next and shuffle it around so that ordinary flash storage can stand in for some of that pricey DRAM, behaving more like the fast stuff. The result is more usable memory at lower cost, without a major hit to speed. For a company running giant AI models, that can mean doing the same work with less of the most expensive and hardest-to-find hardware.
That is why AMD wanted it. Modern data centers are increasingly constrained not by computing power, but by the memory needed to feed it. By folding MEXT’s tools across its lineup, AMD is betting it can offer customers better performance for each dollar they spend — a compelling pitch when a single AI data center can cost billions of dollars to build and equip.
The deal also brings people, not just software. AMD said MEXT’s team has deep experience in memory systems and AI infrastructure, expertise that will help address the engineering challenges of the massive data center buildouts now underway.
The acquisition fits a broader trend reshaping the technology sector. Memory has gone from an afterthought to one of the hottest corners of the AI economy. Companies such as Micron and SanDisk have benefited from surging demand as AI deployment strained global memory supplies and pushed prices higher. AMD’s move takes a different approach: rather than producing more memory, it is investing in technology that helps existing memory go further.
The deal also sharpens AMD’s competition with Nvidia, which continues to dominate the AI-chip market. AMD has spent the past several years positioning itself as a full-service alternative, offering processors, networking, software, and now memory-optimization technology designed to improve data-center efficiency. The same day it announced the MEXT acquisition, AMD also launched a product intended to compete with Nvidia’s DGX Spark platform.
For users, the significance extends beyond a single acquisition. The AI services increasingly used every day — chatbots, image generators, search assistants, and enterprise tools — depend on data centers whose costs continue to climb. Memory is among the most expensive components. Any technology that reduces those costs could influence how quickly AI expands and how much businesses ultimately charge for access.
Whether MEXT’s technology delivers on its promise remains to be seen, and AMD still trails Nvidia by a significant margin in AI hardware. But the acquisition underscores a growing reality in artificial intelligence: success is no longer determined solely by processing power. Increasingly, it depends on solving the memory challenge that sits behind it.
JBizNews Desk
© JBizNews.com All Rights Reserved. Reproduction or distribution without written permission is prohibited.