AMD has acquired memory optimization startup MEXT, bringing predictive memory optimization software into its AI infrastructure portfolio as enterprises look for ways to manage increasingly memory-intensive AI workloads without continually expanding expensive DRAM capacity.

The technology uses AI to intelligently move frequently accessed data between flash storage and DRAM, enabling organizations to increase effective memory capacity while reducing infrastructure cost and power consumption, according to AMD.

“Memory has become a critical constraint across cloud and enterprise environments,” AMD said in a blog announcing the acquisition, adding that traditional approaches of simply adding more DRAM are becoming increasingly costly and power-intensive.

Financial terms of the acquisition were not disclosed. AMD did not immediately respond to a request for additional comment.

AI boom reshapes memory economics

AMD’s move comes as AI infrastructure demand is reshaping the memory market and forcing enterprises to rethink how they scale AI deployments.

According to IDC, AI infrastructure is driving a strategic reallocation of memory production toward enterprise-grade components, with 2026 DRAM supply growth expected to remain below historical norms at 16% year over year, creating pricing pressure across the market.

Gartner has separately forecast a 130% increase in combined DRAM and SSD prices by the end of 2026, warning that higher memory costs will increasingly influence enterprise technology investment decisions.

Against that backdrop, MEXT’s software is designed to address a growing enterprise challenge by using predictive algorithms to identify frequently accessed data and proactively move it between flash storage and DRAM, extending usable memory capacity without requiring proportional hardware expansion, the blog added.

“Memory prices have seen an unprecedented growth, nearly going 4x since 3Q25, making memory one of the most contested chips in the AI infrastructure story,” said Shrish Pant, director analyst at Gartner.

Pant said higher prices and constrained supply are reviving interest in software-driven memory optimization strategies that received little attention when memory was abundant and inexpensive.

AI infra competition moves up the stack

The acquisition also reflects a broader shift in how AI vendors are competing for enterprise workloads.

While the first phase of the AI race centered on securing GPUs and compute capacity, vendors are increasingly investing across networking, software, and infrastructure optimization to improve overall system efficiency.

“We can safely say that we are beyond ‘chips wars’ and have already entered into an ‘Infrastructure optimization war’, and software-based memory optimization is just one of many moving pieces which will determine winners for the AI race,” Pant said.

AMD’s acquisition expands its AI infrastructure portfolio beyond processors into software that optimizes memory utilization, mirroring a broader industry trend toward integrated hardware and software stacks rather than standalone silicon performance.

Manish Rawat, semiconductor analyst at TechInsights, said memory is increasingly becoming a strategic constraint for enterprise AI deployments.

“As enterprises deploy larger models and scale user workloads, memory limitations often constrain performance and GPU utilization before compute resources are fully exhausted,” Rawat said.

Memory is evolving from a supporting hardware component into a strategic enabler of AI scalability, performance, and cost optimization, he added.

Software aims to delay expensive DRAM upgrades

AMD said MEXT’s predictive memory tiering technology intelligently places frequently accessed data in high-speed memory while shifting less active data to lower-cost flash storage.

That approach is intended to increase infrastructure efficiency and reduce the need for continual DRAM expansion as enterprise AI workloads grow, analysts said.

Rawat said software-based memory optimization offers enterprises a practical way to delay expensive hardware upgrades rather than eliminate the need for DRAM.

Although the technology cannot replace high-performance DRAM for latency-sensitive applications, it can improve data center efficiency, lower total cost of ownership, and help organizations maximize returns from existing infrastructure investments, he said.

Sanchit Vir Gogia, chief analyst at Greyhound Research, said the industry is entering a phase where infrastructure orchestration will matter as much as compute performance.

“The GPU is the engine. Memory is the road, the fuel line, and occasionally the traffic jam,” Gogia said.

Production AI workloads place sustained demands on parameters, embeddings, and cached context, making memory performance a business issue rather than simply a hardware specification.

Gogia said predictive memory tiering addresses inefficiencies that often leave expensive DRAM underutilized, but cautioned that optimization should complement, rather than replace, sound infrastructure design.

“Predictive tiering attacks the waste inside that reflex,” he said, referring to the tendency to address performance challenges by purchasing more memory instead of improving utilization.

Rawat said organizations that optimize compute, memory, storage, and software together are likely to scale AI deployments faster, lower operating costs, and generate stronger returns on AI investments than those relying primarily on increasing hardware capacity.