From deep within Chalmers University of Technology in Gothenburg, a unique vision began to take shape: Could AI become not just smarter, but also radically more efficient?
For Hans Salomonsson, co-founder of Embedl, the journey began in a European research consortium tackling a pressing global issue: the vast and growing energy consumption of AI.
”AI is a climate issue. The demand for computing power is skyrocketing and with it, energy use. Our research started with the question: how can we make AI more energy-efficient?”
What followed was the foundation of Embedl in 2018, and a first customer shortly after—an early milestone that proved the technology’s potential in real-world applications. In 2020, that milestone was followed by a successful investment round, enabling the team to accelerate product development and customer outreach.
Making AI Work Smarter
At its core, Embedl develops tools that automatically optimize machine learning models so they run faster and consume less energy. These tools analyze deep learning models and intelligently remove unnecessary components, fine-tune critical parts, and tailor the models to the specific hardware they will run on, whether that’s a car, a drone, a microcontroller, or a robotic arm. The result: Smarter AI that runs better, on less powerful chips, with far lower energy demands.
The technology behind Embedl is, as Hans describes it, “brain surgery for AI models.” Their software goes deep into the architecture of AI systems, removing redundant components, replacing inefficient layers, and optimizing the models to run faster and use less energy, regardless of the hardware.
“AI models aren’t built to be efficient, they’re built to work,” Hans explains. “But if you want AI to run on embedded systems like a car, a drone, or even a household robot, you need to optimize it to the core. That’s where we come in. This hardware-agnostic optimization has led to spectacular results. In industrial applications, we’ve achieved performance improvements ranging from 50% up to 1900%, depending on the system and conditions.”
Reaching New Markets with Scalable Tools
In 2022, Embedl received a €2.5 million grant from the European Innovation Council to take the technology further and make it even more widely available. In 2023, Spintop Ventures led the company’s seed round. To reach a broader developer base, Embedl recently launched a SaaS solution which allows non-experts in AI to embed optimized models into physical devices. It’s a move aimed at accelerating industrial adoption and scaling globally.
Behind the product is a growing team of 25, mostly Chalmers alumni, working out of Gothenburg. Culture, according to Hans, is key: “We hire people who are not only smart, but also kind. We’ve created an environment where people thrive, and that’s essential when you’re doing deep innovation.”
Looking ahead, Embedl sees massive opportunities in sectors like automotive safety, defense automation, and the next generation of robotics, including domestic robots, a future Hans believes is just a few years away.
But to harness these opportunities, he believes Swedish companies need to be bolder:
“Swedish industry is too cautious. If we don’t start investing in real innovation, we risk being left behind. We need to dare more, or the Swedish industry will become irrelevant.”