Dutch Axelera AI raises over $250 million from investors. What exactly does this AI chip company do and why is this relevant for the Dutch manufacturing industry?
Axelera AI was founded in 2021 as a spin-off of the research centre imec. The company focuses on developing specialised AI inference chips that allow artificial intelligence to be implemented directly in devices and systems without dependence on a data centre or the cloud. Based in Eindhoven, the company is led by CEO Fabrizio Del Maffeo. Axelera AI now counts more than 500 customers worldwide, active in sectors ranging from defence and industrial automation to agritech and robotics.
Doing AI calculations in practice
While many tech companies focus on training AI models, Axelera AI specialises in inference: actually performing AI calculations in practice. A key limitation here is the energy consumption of existing hardware such as GPUs, which is often too high for applications in, for example, factories, vehicles and medical devices.
Axelera AI is therefore working on energy-efficient chips for AI inference, also called AI Processing Units (AIPUs). In doing so, it opts for an edge-first architecture, with chips designed to operate within strict energy and thermal constraints. With the Metis and Europa AIPUs, the chip company aims to enable AI inference at a significantly lower cost than a GPU.
Computervision conference on the edge
Metis focuses on computer vision inference on the edge, i.e. the necessary calculations are performed directly on the device. The platform offers up to 214 TOPS of computing power. The platform is scalable, allowing easy expansion of computing power by adding more AIPUs. Europe is stronger than Metis, offering up to 629 TOPS of computing power. With this, the AIPU is targeted at business servers, generative AI applications with multiple users, robotics and automotive infotainment systems, among others.
The technology behind these chips is based on in-memory computing and a dataflow architecture, which make calculations significantly more efficient than with traditional processors. This leads to ten times lower energy consumption than comparable GPU solutions. By processing data locally, Axelera also reduces not only latency, but also privacy risks. After all, data does not need to be transmitted over the internet and is only stored locally.
Predictive maintenance and real-time decision making
For the manufacturing industry, Axelera's technology offers interesting opportunities. Think of predictive maintenance in factories by integrating AI chips directly into machines. This allows repairs to be made before problems arise, reducing downtime and lowering maintenance costs. In logistics and robotics, AIPUs can, among other things, enable real-time decision-making without depending on remote servers.
The latest funding round was led by Innovation Industries and attracted new investors like BlackRock and SiteGround Capital. In addition, existing investors again put money into the AI chip company, including Samsung Catalyst Fund, Invest-NL and the European Innovation Council.
