Products

Developers

Resources

Company

Products

Developers

Industries

Resources

Company

Products

Developers

Industries

Resources

Company

Edge AI shouldn’t force a trade-off.

For years, running AI at the edge meant giving something up: performance for power, or flexibility for cost. Mobilint designs chips built for AI inference so teams no longer have to choose.

The edge used to mean accepting trade-offs.

Edge devices have hard limits on power, heat and space. Edge NPUs were therefore built to maximize cost-efficiency and performance per watt, at the expense of raw performance and programmability.

That worked when edge AI mostly meant CNN-based vision and lightweight models.

Edge devices have hard limits on power, heat and space. Edge NPUs were therefore built to maximize cost-efficiency and performance per watt, at the expense of raw performance and programmability.

That worked when edge AI mostly meant CNN-based vision and lightweight models.

Cost-efficiency

Performance per watt

Performance

Programmability

Accepting the trade-off.

Then the workload changed.

AI is moving into every device

More applications mean tighter constraints.

Models are bigger and multimodal

LLMs and VLMs now need to run on-site, often without internet access.

AI is becoming agentic and physical

Robots and autonomous systems need high-performance inference in the field.

THE EDGE PARADOX

The power, heat and space limits haven’t changed.

Edge NPU no longer can sacrifice performance and programmability.

The power, heat and space limits haven’t changed.

Edge NPU no longer can sacrifice performance and programmability.

Cost-efficiency

Performance per watt

Performance

Programmability

All four, at once.

Built to deliver all four.

Edge NPUs have usually been optimized for efficiency, and data center NPUs for programmability and scale. For more than a decade,

Mobilint has followed one design principle: deliver both. We design the architecture, compiler, software and algorithms together, so that CNNs, LLMs and multimodal models run efficiently on a single chip.

Edge NPUs have usually been optimized for efficiency, and data center NPUs for programmability and scale. For more than a decade,

Hardware–software
co-design

Hardware–software co-design

Hardware and compiler are designed together, so one NPU covers CNN, LLM and multimodal workloads.

Mixed-precision quantization

Each layer gets its own optimal precision, cutting memory traffic by up to 75% while keeping over 99% of original accuracy, with no retraining.

LUT-based nonlinear functions

Lookup tables handle modern activation functions such as GELU and SiLU at very low area and power cost.

490+

validated model architectures

Built on a GPU? It runs on Mobilint.

If your team built and validated a model on a GPU, it can run on Mobilint hardware.

SDK qb supports more than 490 validated model architectures.

If your team built and validated a model on a GPU, it can run on Mobilint hardware.

SDK qb supports more than 490 validated model architectures.

A decade of research before the first chip.

Mobilint’s architecture grew out of AI semiconductor research that began at a KAIST lab in 2014.

The company was founded in 2019 to take that research to production.

ISSCC 2017

DNPU, an 8.1 TOPS/W reconfigurable CNN-RNN processor for general-purpose deep neural networks.

HotChips 2017

An oral presentation alongside Google’s TPU.

Proceedings of the IEEE 2020

A heterogeneous DNN processor with a non-von Neumann architecture.

Since 2014

30+ international papers published

Mobile + Intelligence.

mobile

AI that runs on devices and on-site servers, not only in the cloud.

intelligence

NPUs and software built for inference, from CNNs to LLMs.

mobilint

Efficient AI, wherever it needs to run.

VISION

AI Starts Here.

MISSION

Build Bold, Run Light, Expand.

01. BUILD BOLD

Our own technology, built on trust and expertise

NPUs in multiple form factors with strong performance per watt and cost-efficiency, plus SDK qb

Hands-on developer support

Transparency about technical quality

02. RUN LIGHT

Efficient, sustainable technical support

Lower total cost and long-term sustainability

Lower electricity use and carbon emissions

More secure AI infrastructure

03. EXPAND

Growing our customers’ AI business together

Lower barriers to AI adoption through support and services

Introduce AI to more industries

CORE VALUES

Expertise

Expertise built through continuous research and development

Efficiency

Efficiency that holds up over time

Pioneering

Expanding the AI industry through technical innovation

HISTORY

From a research lab to
building alongside industry innovators.

2026

MLX-A1, CES 2026 Innovation Award (AI Category) Awarded (USA, CTA)

2025

REGULUS, CES 2025 Innovation Award (AI Category) Awarded (USA, CTA)

Selected Ministry of Science and ICT Innovation Premier 1000 Company

Received NIPA ‘NIPA Excellent Partner’ Commendation

2024

Selected for ‘2024 AI+X Top 100’ by the Artificial Intelligence Industry Association

Awarded ‘The 4th AI Industry Award’, Telecommunications Technology Association

Awarded ‘2024 World Smart City Expo Awards’ AI & DATA Category, Ministry of Land, Infrastructure and Transport, Ministry of Science and ICT

Received NIPA ‘NIPA Excellent Partner’ Commendation

CEO Dongjoo Shin ‘AI Industry Development Contribution’ Ministerial Commendation (Ministry of Science and ICT)

2023

Selected for ‘2023 AI+X Top 100’ by the Artificial Intelligence Industry Association

Awarded the 30th Korea ImpaC Tech Awards, Minister of Science and ICT Award

Received 2023 Korea AI Industry Awards, AIIA Chairman’s Award

Global Star Fabless 30 Technology Development Support (Ministry of Trade, Industry and Energy)

Selected for ‘KOREA AI STARTUP 100’ 3 Consecutive Years by KT & Korea Economic Daily

2022

Selected for ‘2022 AI+X Top 100’ by the Artificial Intelligence Industry Association

Selected for ‘KOREA AI STARTUP 100’ 2 Consecutive Years by KT & Korea Economic Daily

2021

Selected for ‘KOREA AI STARTUP 100’ by Korea Economic Daily & KT Economics & Management Research Institute

CEO Dongjoo Shin Awarded the 31st Haedong Young Engineer Award by the Institute of Electronics and Information Engineers

2020

Achieved Korea’s Top Tier in MLPerf, the Global AI Semiconductor Benchmark

Awarded at AI Competitions and Grand Challenges

2019

Mobilint founded

LEADERSHIP

The people behind the architecture

Dongjoo Shin

CEO / CTO

Jeongseung Lee

Chief Algorithm Engineer

Jongjun Park

Chief Compiler Engineer

Seoin Lee

Chief Operating Officer

Sanghyun Yun

Chief Strategy Officer

PARTNERS

Building with the best.

AI Starts Here.
Let’s build what comes next.

Join the team

We’re hiring across NPU architecture, compilers, software

and business development.

We’re hiring across NPU architecture, compilers, software

and business development.

See open roles

Talk to us

Tell us about the model you built on a GPU and where it

needs to run, and we’ll propose the right hardware setup.

Tell us about the model you built on a GPU and where it

needs to run, and we’ll propose the right hardware setup.

Edge AI shouldn’t force a trade-off.

Edge AI shouldn’t force a trade-off.

Edge AI shouldn’t force a trade-off.

The edge used to mean accepting trade-offs.

The edge used to mean accepting trade-offs.

The edge used to mean accepting trade-offs.

Then the workload changed.

Then the workload changed.

Then the workload changed.

AI is moving into every device

Models are bigger and multimodal

AI is becoming agentic and physical

Built to deliver all four.

Built to deliver all four.