MemryX Unveils Scalable Edge AI Platform to Bridge a Critical Gap

ANN ARBOR, Mich. – June 22, 2026 – As the artificial intelligence boom continues to migrate from the cloud to the factory floor, city streets, and consumer devices, a critical chasm has emerged. While AI models have become more powerful, deploying them efficiently in the real world remains a significant hurdle plagued by power constraints, high costs, and fragmented development tools. Addressing this challenge head-on, Ann Arbor-based semiconductor firm MemryX Inc. today unveiled a major expansion of its Cascade platform at the Automate 2026 conference in Chicago.

The company introduced three new hardware products built on its proprietary MX3 accelerator architecture, creating a cohesive ecosystem designed to scale AI applications from low-power prototypes to high-density enterprise servers. This strategic move aims to resolve what the company’s CEO sees as a fundamental industry bottleneck.

"AI doesn't have a model problem anymore, it has a deployment problem," said Ross Jatou, CEO of MemryX, in a statement. "Whether developers and system builders are prototyping on Raspberry Pi, adding AI acceleration to existing systems through USB, or deploying hundreds of video streams on edge servers, they need a common architecture that scales seamlessly."

A Unified Strategy for a Fragmented Market

MemryX's expanded portfolio is a direct answer to the industry's need for versatility and simplicity. The launch introduces a trio of accelerators, each tailored for a distinct stage of the AI deployment lifecycle, yet all unified by the same underlying hardware and software.

The MemryX Cascade 100R Raspberry Pi HAT+ targets the vast and vibrant community of developers, students, and hobbyists. By integrating two MX3 processors, this add-on board brings hardware-accelerated AI to the popular single-board computer, enabling complex machine vision and robotics projects that would be impossible on the Pi’s native CPU. This entry point allows for rapid, low-cost experimentation.

For upgrading existing infrastructure—a common scenario in industrial and commercial settings—the MemryX Cascade 100U USB Accelerator offers a plug-and-play solution. Featuring two MX3 accelerators in a compact USB-C device, it allows businesses to inject powerful AI capabilities into legacy systems without costly and complex hardware redesigns.

At the high end, the MemryX Cascade 100P PCIe Accelerator is engineered for the most demanding enterprise edge environments. This single-slot, passively cooled card packs a 16-chip array of MX3 processors, capable of processing hundreds of real-time video streams in parallel. It’s designed for easy integration into standard servers, maximizing compute density for applications in smart manufacturing, intelligent transportation, and large-scale video analytics.

The strategic brilliance of this lineup lies in the MemryX Developer Hub, a common software environment that supports all three products. This unified toolchain allows developers to compile and deploy AI models built with popular frameworks like PyTorch, TensorFlow, and ONNX across the entire product family. An organization can prototype a concept on a Raspberry Pi, validate it on existing hardware via USB, and scale it to full production on edge servers, all without rewriting code or changing development workflows—a significant competitive advantage in a market often bogged down by disparate tools.

The Power of 'Compute-at-Memory'

Underpinning MemryX's entire strategy is its proprietary 'compute-at-memory' technology. The company's MX3 architecture is a streaming, near-memory dataflow design that fundamentally rethinks how AI calculations are performed. In traditional systems, moving data between memory and processing units—a phenomenon known as the "memory wall"—consumes substantial power and creates performance bottlenecks. MemryX’s design minimizes this data movement by placing compute functions directly adjacent to memory.

This approach delivers remarkable gains in power efficiency. Each MX3 chip, containing 192 compute engines, typically consumes just 0.5 to 3 watts while delivering up to 6 TFLOPS of performance. The company claims its architecture provides up to 20 times better performance-per-watt than conventional AI accelerators in the same class. This efficiency is critical for edge devices where power budgets and thermal constraints are paramount.

Independent analysis appears to support the company's claims of a developer-friendly and efficient platform. A May 2025 report from Berkeley Design Technology, Inc. (BDTI) praised the MX3 M.2 module for its extensive documentation and ease of use, noting it outperformed a more power-hungry and expensive NVIDIA Jetson AGX Orin module in inference speed. This blend of performance, low power, and usability is central to MemryX's value proposition.

Navigating a Competitive Edge Landscape

MemryX is making its move in a fiercely competitive but rapidly growing edge AI hardware market, projected by some analysts to exceed $50 billion by 2030. The company faces established giants like NVIDIA, whose Jetson platform is a dominant force in robotics and autonomous systems, and Intel, with its Movidius chips. It also competes with highly focused startups like Google, whose Coral accelerators are optimized for its TensorFlow framework, and Israel-based Hailo, which also utilizes a dataflow architecture to achieve impressive performance-per-watt.

While some competitors may offer higher peak performance on specific benchmarks, MemryX is positioning itself strategically by focusing on the entire deployment journey. Its key differentiator is not just a single chip's specifications but the holistic, scalable ecosystem that promises to reduce complexity and accelerate time-to-market. By offering a seamless path from a $100 hobbyist board to a multi-thousand-dollar server card, the company is betting that total solution efficiency will win over customers frustrated by the integration challenges of a fragmented market.

From Edge Expansion to Data Center Ambition

This aggressive product expansion is fueled by a solid financial foundation, including $44 million in Series B funding from a consortium of prominent investors like HarbourVest, the NEOM Investment Fund, and the Arm IoT Fund. This capital is not only enabling the rollout of the Cascade platform but is also funding the company's future ambitions.

MemryX has already signaled that its vision extends beyond the edge. The company's roadmap includes the MX4, a next-generation accelerator designed to scale its 'at-memory' architecture into the data center. This move indicates a long-term strategy to challenge incumbents across the full spectrum of AI hardware, from the smallest embedded devices to the largest cloud infrastructure. By proving its technology's efficiency and scalability at the edge, MemryX is building a foundation for a much larger strategic play, aiming to provide a compelling alternative for AI workloads everywhere.