AMD just bought itself a way around one of AI’s most stubborn physics problems. The chipmaker announced its acquisition of MEXT, a startup that uses predictive AI to make flash storage behave more like DRAM, on June 15, 2026.

The deal’s financial terms were not disclosed. But the strategic logic is clear: memory is becoming the chokepoint for AI workloads, and AMD wants to own a fix for it.

What MEXT actually does

Think of it like this. Modern AI models are ravenous for memory. DRAM, the fast memory that processors rely on, is expensive and power-hungry. Flash storage is cheap but slow. MEXT’s software sits in between, using AI to predict which data needs to be in fast memory and which can stay in cold storage, then shuttling bits accordingly.

This addresses what the industry calls the “memory wall,” a well-documented bottleneck where the speed gap between processors and memory limits how large and complex AI models can get.

MEXT’s predictive memory optimization technology is designed to reduce total cost of ownership for enterprise and cloud customers. That matters because DRAM expenses have been climbing steadily as AI workloads demand ever-larger memory footprints.

Why AMD needs this now

AMD plans to integrate MEXT’s technology across its existing portfolio of CPUs and GPUs. That means data-center CPUs, AI accelerators, and general-purpose workloads all stand to benefit.

Market reaction and competitive landscape

AMD’s stock reacted positively following the announcement. Investment analysts noted that the acquisition aligns with AMD’s broader growth objectives in the AI infrastructure market, even without knowing the price tag.