Pat Gelsinger backs Nagare Labs’ Z2RAM bid to break the AI memory wall

Pat Gelsinger spent decades building chips at Intel. Now he is writing checks for the part of the chip that keeps AI waiting.

Playground Global, where the former Intel CEO serves as a General Partner, is backing Nagare Labs, a startup developing an oxide semiconductor memory technology called Z2RAM. The pitch targets the “memory wall,” one of the more stubborn bottlenecks in modern AI hardware.

Nagare Labs is promising a lot. As The Register put it, the road ahead is arduous and littered with the ruins of failed memory vendors.

What Z2RAM claims to do

Z2RAM aims to fix that by combining three traits that rarely show up in the same product. The startup is pitching SRAM-class bandwidth and latency, DRAM-class density, and manufacturing that looks more like NAND flash production.

Each of those memory types is good at one thing and compromises on others. SRAM is very fast and sits close to the processor, but it takes up a lot of chip area. DRAM packs more data into less space, but it is slower.

According to the research findings, Z2RAM is designed to use existing 300 mm manufacturing processes and to eliminate the need for capacitors and interposers.

Conventional DRAM stores bits with tiny capacitors, and high-performance memory setups often rely on interposers, an extra layer of silicon that connects memory stacks to processors. Removing both could, in principle, simplify how these chips get built and assembled.

The technology is built on oxide semiconductors. The approach draws on the founders’ extensive earlier research in that field.

The people behind it

Nagare Labs was founded in 2026 by Suman Datta and Shimeng Yu. The company’s roadmap moves through material selection, functional bit-cell demonstrations, and validation of long-term stability.

Gelsinger’s own path to this deal has been eventful. He retired as Intel’s CEO on December 1, 2024, and joined Playground Global as a General Partner on March 26, 2025.

Why the memory wall matters now

AI workloads are hungry for two things: memory bandwidth and low latency. Bandwidth is how much data can move at once. Latency is how long each request takes to come back.

Nagare Labs is betting that a fresh materials approach can leapfrog incremental improvements to existing memory types.

What this means for the memory market

Nagare Labs’ emphasis on existing 300 mm processes looks like a direct answer to the manufacturing challenge. A memory design that slots into fabrication lines already in operation has a much easier route to production than one requiring exotic new equipment.

The key things to watch are the milestones themselves. Functional bit-cell demonstrations will show whether the core idea works, while long-term stability validation will show whether it can survive outside the lab.

Until those results arrive, Z2RAM remains a promising claim backed by a famous name. Gelsinger’s involvement buys Nagare Labs attention and credibility. Only the silicon can deliver the rest.