MemryX AI Unveils MX3+, Scaling AI Inference Beyond 1 POPS

Next-generation accelerator delivers up to 10× more peak AI compute and nearly 10× more on-chip memory within the same low-power envelope as MX3

ANN ARBOR, Mich., Nov. 10, 2026 /PRNewswire/ -- MemryX AI, a fabless semiconductor company developing highly efficient and scalable AI inference solutions, today unveiled MX3+, its next-generation AI accelerator designed to support larger, more complex models across edge devices, industrial systems, and high-density inference infrastructure.

Building on the production-proven MX3 dataflow architecture, MX3+ is designed to deliver up to 10 times more peak AI compute and nearly 10 times more on-chip memory while remaining within the same low-power operating envelope as MX3. These advancements will broaden the range and complexity of AI workloads MemryX AI can address, including higher-resolution vision, transformer-based models, small language models, vision-language-action models (VLAs), hybrid architectures, and more sophisticated multi-model AI pipelines.

MX3+ will retain the scalable, multi-chip approach established with MX3, allowing multiple accelerators to operate as a unified logical device. Planned configurations will scale from a single accelerator to 16-chip systems capable of delivering more than one peta-operation per second of peak AI compute.

"MX3+ represents a major generational step for MemryX AI—not simply a faster chip, but a platform designed to bring more capable AI into real-world systems while significantly lowering total cost of ownership for high-density edge deployments," said Ross Jatou, President and CEO of MemryX AI. "Customers will be able to use the same architectural foundation across embedded devices, industrial systems, and high-density inference infrastructure, giving them a seamless, highly scalable path to transition complex workloads from development straight to volume production."

Designed to Scale Beyond 1 POPS

MemryX AI's distributed dataflow architecture enables multiple accelerators to work together without requiring the host processor to manage individual chips. The MemryX Neural Compiler automatically maps neural networks across the available accelerators, allowing developers to scale performance without manually partitioning their models.

A 16-chip MX3+ configuration is designed to deliver more than 1 POPS of peak AI compute and 2.16 GB of distributed on-chip memory, supporting deployments ranging from embedded systems to high-density inference infrastructure.

Built for More Advanced AI Workloads

The expanded AI capabilities position MX3+ to address increasingly complex applications across industrial automation, robotics, intelligent transportation, smart infrastructure, video management systems, healthcare devices, retail, and consumer electronics.

Building on MemryX AI's near-memory dataflow architecture, MX3+ combines substantially increased on-chip memory with a pipelined approach that keeps model data close to computation, minimizing data movement. This architecture is designed to support larger models and higher-resolution inputs while maintaining efficient, predictable, low-latency inference at a batch size of one.

MX3+ is also being designed to carry forward the ease of development established with MX3, providing developers with a familiar, straightforward experience for compiling and deploying AI applications, even as model complexity and system scale increase.

Beyond its performance and memory improvements, MX3+ represents an important step in MemryX AI's long-term architecture roadmap. It preserves the modularity, scalability, and distributed dataflow principles established with MX3 while introducing new capabilities that will carry forward into MX4, the company's future platform targeting advanced multimodal and large language model inference.

Availability

MX3+ engineering samples are targeted for late 2027, with volume production targeted for mid-2028. Additional product specifications, system configurations, and availability information will be announced as development progresses.

MemryX AI is currently accepting early-access requests from select strategic partners and enterprise customers interested in co-validating pre-production hardware and software platforms. To request access or schedule an executive briefing, contact [email protected].

About MemryX AI

MemryX AI is a fabless semiconductor company developing highly efficient and scalable AI inference solutions for edge devices, industrial systems, and data-center infrastructure. Its near-memory dataflow architecture is designed to deliver efficient, low-latency inference while allowing developers to retain their existing models, frameworks, applications, and host platforms.

MemryX AI is headquartered in Ann Arbor, Michigan.

For more information, visit www.memryx.ai.

Media Contact
Joe Faris
VP, Sales and Marketing
MemryX AI
[email protected]

SOURCE MemryX