[News] AMD Acquires MEXT to Address AI Memory Bottlenecks with Tech That Makes NAND Act Like DRAM

AMD has acquired MEXT, a startup developing memory optimization technology aimed at easing data center memory bottlenecks. According to Tom’s Hardware, AMD announced on June 15 that it had acquired MEXT, whose memory tiering technology allows NAND flash to function as DRAM from the operating system’s perspective, helping data center operators reduce DRAM costs.

Tom’s Hardware notes that memory resources, rather than CPUs or GPUs, are increasingly becoming the primary performance bottleneck, while DRAM is often underutilized. As the report explains, MEXT addresses this challenge with an AI-driven memory tiering technology that moves infrequently accessed data from expensive DRAM to lower-cost NAND storage.

Its Predictive Memory Engine continuously monitors memory access patterns and uses AI models to predict which data stored in flash will be required next. The system then preemptively moves those memory pages back into DRAM before applications request them, enabling software to access the data as if it were stored in main memory while maintaining performance, the report notes. By increasing the amount of usable memory available to applications, MEXT’s technology can improve infrastructure utilization and reduce dependence on costly DRAM.

Beyond adding new technology capabilities, the acquisition also gives AMD access to a team with specialized expertise in memory architectures, infrastructure software, and large-scale computing systems, Tom’s Hardware notes.

The Next Challenge: Integration and Validation

As a leading supplier of CPUs and GPUs, AMD remains one of NVIDIA’s key challengers in the AI semiconductor market. As noted by Solution News, the acquisition is intended to combine MEXT’s memory optimization technology with AMD’s compute portfolio, enabling the company to offer a more comprehensive data center solution. Meanwhile, validation remains the next major hurdle. Solution News notes that integrating a startup’s technology into a large-scale semiconductor ecosystem and demonstrating the promised performance gains in commercial deployments could prove challenging.

Read more

• [News] AMD Zen 7 Reportedly Built on TSMC A14 Node as Powertech’s FOPLP Packaging Said to Be Under Evaluation
• [News] AMD Lifts Server CPU Outlook to 35%+ CAGR, $120B by 2030; Warns on Memory-Driven PC, Gaming Weakness

(Photo credit: AMD)