MemryX
Fabless semiconductor company developing the MX3 Edge AI Accelerator, a compute-at-memory chip delivering up to 6 TFLOPs at 0.5-2W for power-efficient edge AI inference in industrial, video management, retail, and smart vision applications.
- Company typePrivate
- Founded2019
- HeadquartersAnn Arbor, United States
- Headcount11–50
- GTM typeB2B
- OfferingHardware or Manufacturing
What MemryX does
MemryX is a fabless semiconductor company headquartered in Ann Arbor, Michigan, founded in 2019 by Dr. Wei Lu and Dr. Zhengya Zhang, both professors of Electrical Engineering and Computer Science at the University of Michigan. The company designs and sells the MX3 Edge AI Accelerator, a chip built on proprietary compute-at-memory technology and a configurable dataflow architecture that uses memory as the only interconnect between compute engines. The MX3 delivers up to 6 TFLOPs at 1GHz, consumes 0.5-2W per chip, supports BF16 activations with block floating point for accuracy preservation, and scales from a single chip up to 16 interconnected chips (96 TOPS) using a unified software stack and host interface (PCIe Gen 3 or USB 3).
The product portfolio includes the standalone MX3 chip, the MX3 M.2 AI Accelerator Module (4-chip, 24 TFLOPS, 2280 M-Key form factor, $149 retail price point), and the Cascade deployment platform spanning PCIe, USB, and Raspberry Pi HAT+ form factors. A software suite (Smart compiler, Python/C/C++ APIs, Runtime, Bit Accurate simulator) and the Developer Hub enable customers to compile and deploy their own trained AI models without a model-zoo lock-in, with industrial temperature range support (-40C to +85C). Revenue is generated through hardware sales distributed via direct enterprise sales, channel partners (WPG Americas, Mouser, Macnica), and e-commerce (Amazon), targeting customers in video management systems, industrial automation, retail, automotive, and smart vision applications; named customers include ASUS IoT, DYNICS, Lenovo, ARBOR Technology, DFI, and BIOSTAR.
The company has raised $44 million in cumulative funding across Series A (May 2021, investors including ARM IoT Capital, HarbourVest, M Ventures, eLab Ventures, Motus Ventures, University of Michigan) and Series B (March 2025, $44M). MemryX operates from offices in Ann Arbor, Bangalore, Taipei, Hsinchu, and Riyadh, and was led by Keith Kressin (ex-Qualcomm) as CEO from December 2021 until Ross Jatou (ex-NVIDIA, onsemi, Alat) succeeded him in June 2026 alongside the appointment of Joe Faris (ex-Intel, Luminar) as VP of Sales and Marketing. The company is developing the MX4 next-generation architecture, which extends the compute-at-memory design with 3D memory and asynchronous dataflow targeted at Edge and Data Center applications.
MemryX firmographics
Firmographics- Name
- MemryX
- Legal name
- MemryX Inc.
- Website
- https://memryx.com
- Company type
- Private
- Founded year
- 2019
- Operating status
- Operating
- Headcount range
- 11–50 employees
- Short description
- Fabless semiconductor company developing the MX3 Edge AI Accelerator, a compute-at-memory chip delivering up to 6 TFLOPs at 0.5-2W for power-efficient edge AI inference in industrial, video management, retail, and smart vision applications.
- Ownership category
- akta.pro rank
MemryX industry classification
Industry- Product category
- Edge AI Semiconductor Accelerators
- NAICS
- Navigational, Measuring, Electromedical, and Control Instruments Manufacturing (33451), Computer and Peripheral Equipment Manufacturing (33411)
- SIC
- Industrial Instruments For Measurement, Display, And Control (3823)
- akta.pro primary industry
- Edge AI Hardware & NPUs (SoCs, modules, accelerators) (HDAAAJAC)
- akta.pro secondary industries
- AI Server Systems & HGX/Accelerator Platforms (HDAAAAAB), Advanced SoCs & Application Processors (HDAHAAAB), Embedded & Compute Modules (SoMs, COM Express/SMARC, industrial SBC modules) (IMAGAJAK), AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers) (HDAAAAAI)
Keywords
Where MemryX is headquartered
LocationHeadquarters
- HQ city
- Ann Arbor
- HQ country
- United States
- HQ region
- North America
Offices5 records
Markets served
MemryX business model
Business model- GTM type
- B2B
- Offering type
- Hardware or Manufacturing
- Cost components
- Technology or R&D, Personnel, Supply Chain, Marketing or Sales, Operations
Revenue model
- Hardware Sales - AI Accelerators: Sale of MX3 Edge AI Accelerator chips and M.2 modules as standalone products for edge AI inference acceleration. Products include MX3 chip, MX3 M.2 AI Accelerator Module, and Cascade 100 series (100P PCIe, 100U USB, 100R Raspberry Pi HAT+).
- Enterprise Solutions: Custom enterprise deployments and solutions integration through system integrators and partners for industrial, automotive, retail, and smart city applications.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| One time/ perpetual license | Pay-as-you-go | MX3 M.2 AI Accelerator Module |
| One time/ perpetual license | Pay-as-you-go | MX3 Edge AI Accelerator Chip |
| One time/ perpetual license | Pay-as-you-go | Cascade 100 Series |
Go-to-market motion2 records
Distribution channels5 records
Marketing channels7 records
MemryX product offering
Product offeringCore offering
MemryX designs and sells MX3 edge AI accelerator chips and M.2 modules that deliver high-performance, low-power AI inference for edge devices. The product line has expanded to include the Cascade 100 deployment platform with PCIe, USB, and Raspberry Pi HAT+ form factors, all supported by the MemryX Software Suite (compiler, APIs, runtime) and the Developer Hub. Customers include OEMs and system integrators across industrial automation, video management, retail, automotive, and smart city applications.
Product overview
MemryX is a fabless semiconductor company offering a unified Edge AI accelerator portfolio anchored by the MX3 architecture. The product line includes standalone MX3 chips, 4-chip M.2 modules, and the expanded Cascade platform with PCIe, USB, and Raspberry Pi HAT+ form factors. All products share a common MX3 architecture enabling a single scalable deployment ecosystem from low-power embedded devices to high-density edge servers. The portfolio is complemented by MemryX's software suite (Smart compiler, Python/C/C++ APIs, Runtime) and the Developer Hub for streamlined deployment. The company is developing the next-generation MX4 with 3D Memory for data center applications.
Differentiator
Problem solved
Functional benefit
Brands
- MX3: Edge AI Accelerator chip and M.2 module product line featuring up to 6 TFLOPs per chip with high utilization, 0.5-2W power consumption, and support for up to 16 chips interconnected.
- Cascade
- Developer Hub
Products and services
- MX3 Edge AI Accelerator (Chip) Standalone AI accelerator chip delivering up to 6 TFLOPs at 1GHz with bfloat16 activations and 4/8/16-bit weights. Supports up to 16 chips (96 TOPS) interconnected, consumes 0.5-2W per chip, and is intended for OEMs and system integrators building edge AI products.
- MX3 M.2 AI Accelerator Module 4-chip M.2 module delivering up to 24 TFLOPS at 1GHz in 2280 M-Key form factor (22x80mm). Features 42M 8-bit parameters, x2 or x4 PCIe Gen 3 interface, industrial temperature range (-40C to +85C), and is priced at $149 per unit. Targeted at developers, OEMs, and edge AI deployments requiring high performance in compact form factors.
- Cascade 100P PCIe Accelerator High-density edge server AI accelerator using a PCIe interface, supporting up to 16-chip configurations. Part of the expanded Cascade deployment platform built on the MX3 architecture for scaling AI from development to enterprise edge infrastructure.
- Cascade 100U USB Accelerator USB-based AI accelerator for rapid AI prototyping and quick deployment and testing of AI models. Part of the Cascade deployment platform built on the MX3 architecture.
- Cascade 100R Raspberry Pi HAT+ Low-power AI accelerator in Raspberry Pi HAT+ form factor for industrial applications. Enables AI deployment on compact single-board computing platforms, built on the MX3 architecture.
- MemryX Software Suite Comprehensive software stack including the Smart compiler for automated AI mapping, Python and C/C++ APIs, Runtime with driver and firmware for Windows/Linux, and a Bit Accurate simulator for testing models without hardware. Enables users to install and begin compiling AI models in under one hour.
Quantifiable outcome
- Uses less than 10% power of mainstream GPU solutions
- +5 more outcomes
Companies that use MemryX
Customer profileNamed customers6 records
Segments6 records
Ideal customer profiles6 records
MemryX technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
Integration16 records
AI capability4 records
Feature6 records
MemryX partnerships and signals
Strategic signalPartnerships
14 partnerships are on record, tiered core, flagship and minor.
- Macnica, Inc.coreDistribution and solutions partnership with Macnica, one of Japan's largest semiconductor distributors. Macnica became MemryX's primary channel partner in Japan offering localized technical support to expand access to AI accelerators in the Japanese market.
- DataturecoreStrategic partnership to develop advanced Vision AI solutions at the edge, integrating ARBOR's platforms, Datature's no-code Vision AI software, and MemryX's AI accelerator chips for rapid deployment.
- National Semiconductor Hub of Saudi ArabiaflagshipStrategic partnership bringing increased development and customer support within Saudi Arabia, expanding MemryX engineering and customer reach across the Middle East.
- NEOM Investment FundflagshipPartnership to accelerate development and deployment of next-generation, energy-efficient edge AI accelerators critical for NEOM's intelligent digital infrastructure.
- Edge AI FoundationminorStrategic partnership supporting the mission to democratize and advance edge AI technologies through scalable, high-performance solutions.
- VariscitecoreStrategic partnership combining MemryX's AI acceleration technology with Variscite's System on Module (SoM) solutions for enhanced performance and efficiency in industrial and edge computing applications.
- DFIcoreHardware partner integration of MemryX AI accelerators into DFI's industrial edge AI platforms including ECX700-ADP rugged system for PPE detection.
- BIOSTARcoreHardware partner integrating MemryX accelerators into BIOSTAR's EdgeComp product line for smart manufacturing, smart retail, and smart city applications.
- ARBOR TechnologycoreStrategic partnership combining ARBOR's edge computing platforms with MemryX AI accelerator chips for industrial Vision AI solutions. Includes airport baggage handling systems (ARES-1983H-AI).
- Cognitica AIflagshipStrategic partnership to create flexible, scalable, and efficient Edge-AI platform for large-scale, high-throughput systems with extensive experience in edge deployment.
- CadencecoreTechnology partnership for chip design tools, with Cadence tools used since MemryX's inception for development of edge AI processing solutions.
- Edge ImpulsecoreSoftware partnership integrating Edge Impulse's AI developer platform with MemryX hardware to streamline development, testing, and deployment of AI models.
- Nota.aicoreStrategic partnership enabling Nota.ai's software platform to search, compress, and automatically deploy AI models on MemryX edge AI processors.
- DeGirumcoreMemryX supported as hardware on DeGirum AI Hub, enabling instant inference using models for object detection, instance segmentation, pose detection, and FPS benchmarks.
Scale indicators9 records
Recent moves8 records
Expansion highlights8 records
MemryX competitors and assessment
Company assessmentEmerging players
- Ambarella: Edge AI SoC company specializing in low-power computer vision and video processing chips for automotive, security, and IoT cameras. Comparable to MemryX in serving VMS and edge vision workloads with integrated AI accelerators, though with a more video-SoC-centric architecture.
Broad incumbents
- NVIDIA: Dominant GPU and AI accelerator supplier with the Jetson edge AI platform (Nano, Orin, Thor) directly competing in edge AI. MemryX's MX3 is positioned as a lower-power, lower-cost alternative to Jetson modules for vision workloads; NVIDIA's incumbent status and CUDA ecosystem make it the primary competitive threat.
- Intel: Semiconductor giant with Movidius VPU and OpenVINO toolkit for edge AI inference, plus broader data center and embedded offerings. Competes with MemryX at the edge through Movidius hardware and SDK-based deployment, leveraging massive scale and ecosystem.
- Qualcomm: Mobile and edge AI silicon provider with AI Engine across Snapdragon platforms. Competes with MemryX in on-device AI inference for IoT, automotive, and edge compute; MemryX's former CEO Kressin came from Qualcomm's Compute and Cloud division.
- Google: Provider of Edge TPU and Coral devices for on-device ML inference. Direct overlap with MemryX's MX3 in the edge AI accelerator category, backed by Google's TensorFlow ecosystem.
Direct peers
- Blaize: Edge AI inference chip company offering Graph Streaming Processor (GSP) architecture for automotive, retail, and industrial edge deployments. Competes head-to-head with MemryX in low-power edge AI accelerator silicon.
- Syntiant: Developer of ultra-low-power edge AI processors (NDP120, NDP200) for always-on audio and sensor applications. Direct peer in the edge AI accelerator space targeting low-power inference at the edge, with overlapping industrial and IoT customers.
- Sima.ai: Edge AI chip company offering MLSoC accelerators for embedded vision systems. Comparable in targeting low-power vision AI at the edge with similar form factors and developer-focused software toolchains.
- Mythic: AI processor company using analog compute-in-memory (CIM) technology for edge AI inference. Closely comparable to MemryX because both leverage in-memory compute approaches for low-power edge AI; Mythic focuses on vision AI workloads similar to MX3.
- Hailo: Israeli fabless AI chip company building edge AI accelerators (Hailo-8, Hailo-15) for vision and ML inference at the edge. Directly competes with MemryX MX3 in the same edge AI accelerator category with similar price/performance positioning and OEM go-to-market.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat5 records
Key risks6 records
Key highlights6 records
Customer concentration
MemryX social profiles
Digital presenceMemryX financial estimates
Financial estimateRevenue estimate
Valuation estimate
MemryX leadership team
Management profileNumber of profiles
Profiles10 records
MemryX funding detail
Funding detailFunding overview
Funding rounds4 records
Investors8 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
MemryX M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about MemryX
What does MemryX do?
MemryX designs and sells MX3 edge AI accelerator chips and M.2 modules that deliver high-performance, low-power AI inference for edge devices. The product line has expanded to include the Cascade 100 deployment platform with PCIe, USB, and Raspberry Pi HAT+ form factors, all supported by the MemryX Software Suite (compiler, APIs, runtime) and the Developer Hub. Customers include OEMs and system integrators across industrial automation, video management, retail, automotive, and smart city applications.
Is MemryX a public or private company?
MemryX is a private company. It is classified as venture growth investor backed and is currently operating.
When was MemryX founded?
MemryX was founded in 2019. It employs 11 to 50 people.
Where is MemryX based?
MemryX is headquartered in Ann Arbor, United States, in the North America region.
How does MemryX make money?
Two revenue lines are on record. Hardware Sales - AI Accelerators are the primary driver. The others are enterprise Solutions.
Who are MemryX's main competitors?
Ambarella is listed as an emerging player. Broad incumbents are NVIDIA, Intel, Qualcomm and Google. Direct peers are Blaize, Syntiant, Sima.ai, Mythic and Hailo.
Does MemryX have an API?
Yes. MemryX provides Python and C/C++ low and mid-level APIs for AI integration. The Developer Hub offers comprehensive software tools including a Smart compiler for optimized AI mapping, powerful APIs, Runtime (driver and firmware for Windows or Linux), and a Bit Accurate simulator for testing models without hardware. Developer documentation is at developer.memryx.com/docs.
What industry is MemryX in?
MemryX's product category is Edge AI Semiconductor Accelerators. Its primary akta.pro industry code is HDAAAJAC, Edge AI Hardware & NPUs (SoCs, modules, accelerators), with a secondary code of HDAAAAAB, AI Server Systems & HGX/Accelerator Platforms. Its NAICS code is 33451 and its SIC code is 3823.