AI that runs without servers, data centers, or sky-high energy costs? Meet Embedl, the Swedish deeptech startup building the AI layer your cloud provider definitely doesn’t want you to know about.
Embedl, a spin-out from Chalmers University of Technology, just raised €5.5 million in pre-Series A funding to fuel what could be the next big shift in AI: running it at the edge, on the actual devices we use in the real world, far from the cloud.
Led by Chalmers Ventures, the round also included Fairpoint Capital, SEB Greentech, Spintop Ventures, and STOAF, bringing Embedl’s total funding to €12 million.
Until now, only AI pros could optimise for edge. With the Hub, Embedl is bringing that power to every developer and product team. The company is building core AI infrastructure for sectors like defense, robotics, and automotive—industries under mounting pressure to deploy smarter systems without ballooning costs or energy footprints.
At the heart of its mission is “edge AI inference optimisation”—a once-niche field now racing into the spotlight. As global demand for artificial intelligence continues to surge, so too does the need for models that can run quickly and efficiently on local hardware, without relying on cloud connectivity or high-power data centers.
“The world needs to make AI more energy efficient—fast,” said Hans Salomonsson, Embedl’s CEO and co-founder. “AI applications are exploding, but we can’t let energy consumption explode with them. Our technology makes AI viable for mass deployment in physical products.”
From university lab to industry workhorse
Founded in 2022 and based on research by Professor Devdatt Dubhashi in data science at Chalmers, Embedl’s tools help companies optimise and deploy AI models for devices like autonomous vehicles, drones, industrial robots, and surveillance systems.
Its core products, the Model Optimisation SDK and the soon-to-be-launched Embedl Hub, are designed to shrink models, cut power usage, and dramatically reduce inference time. In real-world use cases, Embedl claims up to 83% energy savings, 95% memory reductions, and 18x faster inference.
That’s especially important in industries where devices must function independently—far from cloud access, in real-time, and often under tight energy constraints.
Edge AI: From hype to competitive edge
“Edge AI” refers to the execution of machine learning models locally on hardware devices—rather than in a centralised data center. In 2024, inference costs surpassed training costs for AI models globally, a tipping point that is forcing companies to rethink how and where they run AI workloads.
Embedl is positioning itself as the efficiency layer for AI at the edge, a space now seen as one of the most commercially urgent in AI deployment.
From Bosch to Kodiak Robotics, Embedl’s existing clients already use the SDK to optimise complex deep learning models for use in real-world devices, ranging from trucks to wearable sensors. Shubham Shrivastava, Head of Machine Learning at Kodiak, called the technology “game-changing” for its ability to “inspect cognitive blocks,” tailor performance to hardware, and deploy across platforms.
Funding to fuel wider adoption
The new capital will support Embedl’s commercial rollout of its SaaS platform, Embedl Hub, aimed at making edge-AI deployment accessible even to teams without deep AI expertise. While the SDK targets advanced users, the Hub will serve a broader developer base, including those embedding AI into everyday products.
It’s a bet on the broader trend: edge AI is quickly moving from R&D to a must-have for cost-conscious OEMs.
“This funding is a sign that Chalmers has the technical expertise to build great AI solutions,” said Jonas Bergman, Investment Director at Chalmers Ventures. “We expect great things from Embedl—this is just the beginning.”
Going global, staying lean
With a team that’s still under 20 people and 23% women, Embedl is competing not just with larger players, but with a pervasive industry mindset: that internal AI optimisation is always better than outsourced tools.
“Our biggest competitor is the ‘not invented here’ syndrome,” said Salomonsson. “But once teams see our benchmarks and results, we quickly earn their trust.”
The startup isn’t disclosing its current valuation, but its goals are clear: over the next three to five years, Embedl aims to lead the edge-AI space, expanding its product portfolio while targeting industries that need to run AI without excess compute.
And with global AI energy usage skyrocketing—and edge devices multiplying across sectors—it may be right on time.
