From research idea to business – Embedl wants to make AI ready for the real world
As artificial intelligence moves out of the cloud and into cars, cameras and industrial systems, the rules of the game are changing. It is no longer enough for models to be accurate – they also need to be fast, energy-efficient and able to operate in real time. That is where Embedl comes in.
The Gothenburg-based company Embedl has developed technology that makes it possible to run AI directly inside a product, without the need for a cloud connection. The background is a fundamental challenge in modern AI: larger and more advanced models often deliver higher accuracy, but at the same time require more computing power, more energy and become more expensive to run.
Embedl’s platform first analyses how an AI model is structured and which parts are most important. The model is then compressed by removing unnecessary components.
“An AI model consists of billions of calculations in a computational graph. What we do is go in with our algorithms and perform ‘brain surgery’ on those calculations. We remove a lot, replace certain parts and reshape the model so that it becomes efficient on the hardware it is actually meant to run on,” says co-founder and CEO Hans Salomonsson.
The results of this brain surgery are often dramatic. Energy consumption is reduced by between 20 and 90 per cent, depending on the application, while costs are cut by 40 to 90 per cent.
Embedl started off as a research project at Chalmers University of Technology, where Hans and Devdatt Dubhashi in 2017 started testing different ways of improving energy efficiency in AI.
“Our first research project was in low-energy computing, an EU-funded project. We saw many opportunities. Then we tested our technology across different use cases—from nuclear physics to chip design and language models.”
When the company took shape in 2020, the focus narrowed, not least because of its proximity to the automotive industry.
“Because we are in the ‘Detroit of Sweden’, close to Volvo, the automotive sector was the natural starting point, followed by defence and security, and more recently robotics.”
A platform covering more steps in the process
Hans Samuelsson explains that Embedl initially saw model optimisation – with streamlined AI models – as the main industrial challenge. But discussions with customers changed that view.
“Optimisation is important, but it is not the only difficult part. Just getting an AI model to run in an embedded system can be very challenging.”
This led Embedl to broaden its solution. Today, the company is developing a platform that supports several stages of the development process, from adaptation and testing to integration and collaboration between different teams.
Its customers operate where requirements are highest, primarily in the automotive industry and defence. They include Bosch, BMW and Kodiak. The technology enables, for example, a car to detect obstacles without delay, a camera to analyse video locally, and a machine to detect faults without sending data externally.
Embedl currently has 32 employees, many with backgrounds in engineering and physics, and several with doctoral degrees. The link to academia remains strong, including guest lectures and collaboration with Chalmers. Asked how the company finds talented employees, the answer is: “They come to us.”
“The most capable people want to work on the hardest problems, and because we tackle those kinds of challenges, we manage to attract very strong talent.”
Alongside product development, Embedl has also changed how it reaches users.
“We have shifted away from traditional business-to-business sales at conferences and trade fairs – they are expensive and you only get a handful of leads each time. Going digital is a completely different scale. Instead of talking to 20 people at a conference, 25,000 may actually evaluate and test our model. But the internet is noisy, so we need to cut through that,” says Hans Salomonsson.
The digital strategy has also highlighted a classic challenge in the AI world: competition from free alternatives.
“There is a huge amount of open-source software available at no cost. That means you have to convince very skilled engineers why they should not build it themselves, but instead use our solution,” says Hans Salomonsson.
The challenge was not that the technology lacked value, but that it clashed with culture. Many engineers are used to building their own tools and place a high value on independence. Paying for something that could be built in-house – even if it takes longer – can be a hard sell.
That insight became a turning point for the company.
“At that point, you are no longer selling to engineers’ development budgets. It becomes a question for managers, or their managers,” says Hans Salomonsson.
SEB Greentech Venture Capital as a catalyst
In 2025, SEB Greentech Venture Capital invested in Embedl.
“The investment allowed us to continue hiring while also driving the commercialisation process. In addition to capital, we have received support with structure, corporate governance and customer introductions. That has been important in the journey from a research company to a business that needs to operate commercially.”
Asked about role models, Hans Salomonsson takes a different path from many others in the tech world.
“Steve Jobs has never been a role model for me, and Elon Musk is difficult to relate to nowadays. Instead, I am inspired by Nvidia’s chief executive officer Jensen Huang, who has managed to combine deep technical expertise with a clear vision.”
However, his main source of inspiration comes from within the company.
“I am most inspired by our engineers. When they quickly put something together and demonstrate a new idea, it gives enormous energy. As a CEO, it is easy to get caught up in meetings, follow-ups and problems that need solving. That’s why it’s valuable to occasionally step back and talk about technology to recharge.”
Read more about Embedl at: embedl.com
Article published on May 5, 2026
Text: Malin Edwards
Photo: Louise Nylund