US-based semiconductor firm AMD has announced the acquisition of MEXT, which specialises in AI-driven memory optimisation technology.
The financial terms of the deal were not disclosed.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
The move comes as data centre operators face mounting challenges around memory access. These pressures are growing particularly as the size and complexity of AI models, analytics, and high-performance computing workloads continue to increase.
MEXT, which launched its Predictive Memory software in April, has developed technology intended to enable flash storage to function more like DRAM.
The firm’s AI-powered approach automatically moves memory pages that are not actively used to flash, reducing reliance on the more costly DRAM resource.
The software predicts which offloaded pages will soon be needed and returns them to DRAM proactively, with the aim of maintaining application performance while expanding effective memory capacity.
According to AMD’s senior vice president and compute and enterprise AI general manager Dan McNamara, the integration of MEXT’s technology is expected to make a difference for customers working in cloud and enterprise environments, where memory remains a critical constraint.
McNamara, in a blog on AMD, wrote further that the inability to access adequate memory can limit improvements to performance per dollar, operational efficiency, and the pace of large-scale deployments.
The newly acquired technology offers a software-only solution that, as described by MEXT, can increase usable memory capacity by two to four times and lower infrastructure costs by up to 50%.
MEXT also claims no changes are required to existing hardware, operating systems, or applications, and the software runs on a single CPU core without needing a GPU.
AMD stated that integrating MEXT’s technology into its data centre portfolio is expected to enhance the value enterprise customers gain from their infrastructure investments and speed up AI deployment.
The deal also sees AMD adding MEXT’s team, known for its expertise in memory systems and AI infrastructure.
AMD described the acquisition as a step towards enabling customers to deploy computing workloads at improved efficiency, cost, and scale.
Demand for memory resources in data centres has risen sharply due to advances in AI infrastructure, driving up costs for critical components such as DRAM.
MEXT’s Predictive Memory works both on-premises and in the cloud, offering predictions in microseconds and continuous self-optimisation.
Last month, OneQode entered into a partnership with AMD to use AMD Instinct GPUs and to implement the AMD Helios rack-scale solution as part of its international AI infrastructure expansion.
