Company profile
MemryX
MemryX develops edge-AI accelerator chips and modules for low-power inference in embedded systems, servers and multi-camera deployments. Its Cascade expans...
Facts
- Segment
- AI_COMPUTE
- Country
- US
- Headquarters
- Ann Arbor, Michigan
- Founded
- 2019
- Categories
- Components
- Focus
- Edge AI compute
- Flagship
- MX3 / Cascade
- Website
- https://memryx.com/
About MemryX
MemryX’s MX3 accelerator uses an at-memory dataflow architecture that distributes neural-network operations across 192 compute engines. A single chip provides approximately 6 TFLOPS while operating at roughly 0.5–3 W, and as many as 16 devices can be cascaded for larger models or multiple concurrent video streams. The compiler accepts established models without requiring retraining or specialised quantisation work.
Cascade products package those chips into PCIe cards, USB modules, M.2 devices and a Raspberry Pi HAT+, allowing developers to add inference to x86, Arm or RISC-V hosts. The CPU continues to run the robot or application software while MemryX executes vision models for detection, inspection, safety and navigation. This division is intended to keep power and host utilisation predictable as camera count or model complexity increases.
Market context
SUPPLY_CHAIN
Summary
Type: AI compute supplier
Core: MX3 / Cascade
Function: Run models
Audience: Robot OEMs
Research, Articles and recent verified events
- Ebttikar and MemryX target Saudi computer-vision deployments — 2026-07-09
- MemryX expands Cascade edge-AI accelerator platform — 2026-06-22
- MX4 roadmap introduces distributed asynchronous dataflow design — 2025-12-26
- Macnica becomes MemryX distribution and solutions partner in Japan — 2025-12-19
- MemryX announces Riyadh office and local engineering teams — 2025-09-25
- MemryX joins Saudi National Semiconductor Hub — 2025-05-06
- NEOM Investment Fund partners with MemryX on edge AI — 2025-05-05
- MX3 M.2 wins 2025 Edge AI Vision product award — 2025-04-21