Around Brainport, a cluster of chip startups is taking on the battle for efficient AI inference

While AMD and Intel pour billions into architectures for large-scale AI training and data centres, a number of Dutch startups are seeking space in a different part of the market: energy-efficient inference at the edge of the network. Most of these companies are located in or around Eindhoven, on and around the High Tech Campus, which has provided an ecosystem for semiconductor companies and suppliers for decades.

The companies vary considerably in their approach: from chips for computer vision to brain-inspired sensor processing and quantum processors. What they share is a preference for specialised architectures over general-purpose hardware. That choice demands significant capital and staying power, but also offers a distinctive position relative to broader-based manufacturers.

Axelera AI is building a full chip roadmap

Axelera AI, founded in 2021 as a spin-off from research institute imec, is the most heavily funded company in this cluster. It develops chips based on digital in-memory computing, a technique in which computations take place at the location where data is stored, sharply reducing memory traffic and, with it, energy consumption.

The first commercial chip, Metis, delivers up to 214 TOPS at an efficiency of around 15 TOPS per watt and is suited for applications such as computer vision and robotics. Models of up to eight billion parameters fit on a single chip. Its successor, Europa, targets edge servers and enterprise systems, offers up to 629 TOPS, and features eight AI cores with a second-generation D-IMC architecture complemented by RISC-V vector cores. Deliveries of Europa-based PCIe accelerators are expected in the first half of 2026. A third chiplet, Titania, is intended for data-centre scale and is planned for 2027.

Axelera AI has raised substantial capital: approximately $450 million in total across equity, loans, and grants. In February 2026 it closed a $250 million Series C round led by Innovation Industries, with BlackRock and SiteGround Capital joining as new investors alongside existing backers including the European Investment Council Fund, Samsung Catalyst Fund, and Invest-NL. Earlier, in March 2025, the company received an EU grant of €61.6 million through the EuroHPC Joint Undertaking DARE project for the development of Titania. This underscores that public and private investors regard the approach as strategically relevant to European digital infrastructure.

Innatera targets sub-milliwatt power consumption

Innatera, founded in 2018 and based in Rijswijk, takes a different approach. The company develops neuromorphic chips, designed on the principle of biological neurons, for sensor data processing. The aim is inference below one milliwatt, a threshold relevant for wearable devices, medical sensors, and industrial IoT applications with tight energy budgets.

The company has raised €19 million to date in a Series A round. Neuromorphic chips remain a relatively early-stage segment: the technology promises very low latency and energy consumption, but also requires adapted programming models that differ from conventional ML frameworks. Innatera is working on tooling to close that gap.

Sensorfy and BQrious address the software and data side

Not every company in this landscape builds its own silicon. Sensorfy, founded in 2009 and based in Eindhoven, combines IoT hardware with software for predictive maintenance in industry. It raised €45 million in a Series C round in 2019. The focus is on translating sensor data into actionable insights for factories and installations, with proprietary hardware and algorithms working in concert.

BQrious Engineering, based in Roermond and founded in 2022, focuses on data science and process optimisation for industrial digital transformation. It received €3.3 million in seed funding in 2025. The company does not develop its own chips, but builds on existing hardware to make industrial processes smarter.

These companies illustrate that the ecosystem is broader than chip design alone: they provide what is needed for the step from raw inference capacity to deployable industrial applications.

QuantWare operates in a separate segment

QuantWare, founded in 2021 and based in Delft, sits outside the AI inference chain but belongs to the wider hardware cluster. The company supplies scalable superconducting quantum processors to quantum computer builders worldwide. In January 2025 it raised €27 million in an extended Series A round. Quantum hardware and AI chips currently share little technological overlap, but both require a similar combination of deep materials science research, manufacturing expertise, and capital-intensive development trajectories.

Position relative to major manufacturers

AMD and Intel are investing heavily in inference processors, including the AMD Instinct series and Intel Gaudi, but these are aimed primarily at data centres with ample cooling capacity and energy budgets. The Dutch startups are seeking their position in applications where wattage and form factor are limiting factors: factory floors, cameras, medical devices, autonomous sensors.

That segment is less mature and less occupied by large manufacturers, although competition is growing. Companies such as Hailo (Israel) and Kneron (Taiwan/US) operate in comparable niches. Whether Axelera AI, Innatera, and their Eindhoven neighbours can hold their ground depends on production scale, the ecosystem around their SDKs, and the pace at which customers deploy edge inference at scale.

For the Dutch and European AI scene, the cluster around Brainport is a tangible indicator that chip design expertise is not concentrated solely at established names. The combination of university research, imec networks, the ASML supply chain, and available venture capital is attracting deeptech teams. Whether this will eventually give rise to an independent European chip industry beyond the manufacturing sphere of TSMC and Samsung remains an open question, but the groundwork is already being laid in Eindhoven and Delft.