Where is MemryX headquartered?

MemryX is headquartered in Ann Arbor, Michigan, USA.

MemryX operates in the AI Chips / Semiconductors sector. Fabless semiconductor company building low-power at-memory dataflow AI accelerator chips (MX3, MX3+, MX4) for edge and data center inference. Efficient, scalable, easy-to-deploy AI inference acceleration that cuts data movement. Customers keep their existing trained models and frameworks and compile them to the accelerator with the MemryX toolchain.

When was MemryX founded?

MemryX was founded in 2019. It is based in Ann Arbor, Michigan, USA.

Who is the CEO of MemryX?

Ross Jatou is the CEO of MemryX. The company was founded in 2019.

Where is MemryX located?

MemryX is headquartered in Ann Arbor, Michigan, USA.

What funding stage is MemryX at?

MemryX is currently at the Series B stage.

MemryX

Category: AI Chips / Semiconductors

Fabless semiconductor company building low-power at-memory dataflow AI accelerator chips (MX3, MX3+, MX4) for edge and data center inference. MemryX was founded in 2019. The company is led by Ross Jatou. Based in Ann Arbor, Michigan, USA. Total funding raised: $50.5M. Latest round: Series B. Key investors include HarbourVest Partners, NEOM Investment Fund, Arm IoT Fund, eLab Ventures, M Ventures, Motus Ventures, Innovation Partnerships.

Founded: 2019
Headquarters: Ann Arbor, Michigan, USA
Total funding: $50.5M

Value proposition

Efficient, scalable, easy-to-deploy AI inference acceleration that cuts data movement. Customers keep their existing trained models and frameworks and compile them to the accelerator with the MemryX toolchain.

Products and solutions

MX3 edge AI accelerator (in production, 6.25 TFLOPS per chip, 10.5 MB on-chip weight memory, 200 mW-3 W), MX3+ (in design, announced 2026-11-10, targets up to 10x peak compute and ~10x on-chip memory, scales to 16 chips as one logical device, engineering samples late 2027, production mid-2028), MX4 (planned, 2029, multimodal/generative AI and LLM), Cascade 100M (M.2), 100P (PCIe), 100R (Raspberry Pi 5 HAT+) and 100U (USB) modules, MemryX Neural Compiler, SDK and Developer Hub, server platforms.

Unique value

Dataflow at-memory compute architecture that minimizes data movement, delivering low-power inference with predictable low latency at batch size 1. The same architecture scales from embedded devices to multi-chip systems.

Target customer

OEMs, system integrators and embedded/industrial product makers building edge AI products, plus emerging high-density edge and server deployments.

Industries served

Industrial automation and robotics, video management systems, transportation/ITS and smart infrastructure, smart and consumer devices, medical devices, retail.

Technology advantage

Proprietary compute-at-memory dataflow architecture proven across MX1, MX2 and MX3. The Neural Compiler automatically partitions models across up to 16 cascaded chips and accepts ONNX, TensorFlow Lite, Keras and TensorFlow models. The toolchain is a free download and the company publishes open-source drivers and demos on GitHub. Won the 2025 Edge AI and Vision Product of the Year Award (Cascade 100M).

How they differentiate

Compute-at-memory dataflow architecture with a compiler-driven, near-automatic multi-chip scaling path (1 to 16 chips) and a consistent roadmap from edge to data center. Third-party testing has praised its ease of use and low power.

Main competitors

Hailo, Axelera AI, DEEPX

Key partnerships

Hardware partners listed on its partner page include Advantech, ASUS IoT, Lenovo, Supermicro, Lanner, DFI, IEI, Aetina, Variscite and Telit Cinterion, also software partners and 6 distributors. Arm IoT Fund is both an investor and an ecosystem tie. University of Michigan is the technology origin.

Major milestones

2019 founded in Ann Arbor, 2021 MX1 silicon and Series A, 2023 MX2, Mar 2025 $44M Series B, 2025 MX3 production and Edge AI and Vision Product of the Year Award, 2026-06 Ross Jatou becomes CEO, 2026-11-10 MX3+ announced (engineering samples late 2027, production mid-2028).

Growth metrics

MX3 in volume production with more than a dozen customer design-ins (per co-founder post, Mar 2025). Four Cascade form factors shipping through distributors; 42 hardware, software and distributor partners listed.

Market positioning

Fabless edge AI inference accelerator vendor, positioned as the most power-efficient option for vision and physical AI, and moving toward generative AI and LLM workloads with MX3+ and MX4.

Geographic focus

Global, with engineering in Ann Arbor (US) and Hsinchu (Taiwan) and operations in Bangalore; distribution through partners in the USA, Japan and the Americas.

Patents and IP

Patent-protected memory processing unit technology; the VP of Architecture alone holds 17 granted US patents (company-wide patent count not verified).

About Ross Jatou

Ex-NVIDIA VP of Engineering (15 years; AI, enterprise, automotive and embedded); ex-onsemi SVP/GM for sensors and processors and President of Alat (Feb 2024-Jun 2026); ex-IBM (2 years). Appointed MemryX CEO on 2026-06-01, succeeding Keith Kressin.

Latest news about MemryX

MemryX targets 10-fold performance gain with MX3+ edge AI accelerator

More AI Chips / Semiconductors companies

Official website: https://memryx.com