Europe's Inference Play: Axelera Launches Europa, Signs AI Factory Deals Worth Tens of Millions

Europe’s most serious attempt yet to break into AI inference silicon moved from slideware to signed contracts this week. On Tuesday, September 15, Dutch chipmaking startup Axelera AI announced that it has signed multiple contracts to supply chips to AI factories, timed with the commercial launch of Europa, its second-generation AI processor. The chips will ship inside specified Dell and Supermicro products — a distribution footprint that almost no European AI chip startup has ever secured.

What happened

Amsterdam-based Axelera, founded in Eindhoven in 2021 as a spin-out from imec, the Belgian nanoelectronics research institute, said the Europa launch is accompanied by signed deals “worth tens of millions” of dollars, according to CEO Fabrizio Del Maffeo, who spoke to Reuters by phone. He declined to give precise contract figures but said the company now serves more than 600 customers running Axelera silicon in security, defence, AI factories, and enterprise settings.

Beyond booked business, the company is pursuing roughly $1.5 billion in potential sales — a pipeline figure Del Maffeo was careful to caveat: it represents opportunities in negotiation, not orders or revenue. Still, for a startup that has raised just over $450 million since 2021, the gap between its balance sheet and its pipeline tells you how sharply European demand for local inference capacity has inflected.

Inside Europa: 629 TOPS at 45 watts

Europa is Axelera’s second-generation AI Processing Unit (AIPU), and it marks a deliberate move upmarket from the company’s Metis line. Where Metis targeted “edge” deployments — a camera analysing security footage on a factory floor, for instance — Europa is built for heavier jobs: enterprise servers, multi-user generative AI, robotics, and automotive infotainment systems.

The headline specifications:

• 8 second-generation AI processing cores delivering 629 TOPS (trillions of operations per second) at a 45-watt TDP
• 16 RISC-V vector processing cores handling fully on-board pre- and post-processing, which frees the host CPU for business logic
• 128 MB of L2 SRAM and 200 GB/s of DRAM bandwidth, with 64 GB of memory capacity on the board
• Native support for INT4/INT8/INT16 precision, plus expanded state-of-the-art model coverage

The efficiency story is the pitch: roughly 14 TOPS per watt in a package designed to run full generative-AI workloads locally, without datacenter infrastructure. That positioning — serious inference performance at edge-adjacent power envelopes — is precisely the gap Nvidia’s mainstream accelerators leave open, and where rivals from Qualcomm to a crowded field of European startups are pushing.

The EU AI factory connection

The contracts are not happening in a vacuum. The European Union is in the middle of revamping its network of supercomputing centres into dedicated AI factories, an effort to narrow the compute gap with the US and China and guarantee that European firms and scientists have access to AI computing power on-shore.

Axelera is working with Dell and systems integrator E4 as one of the suppliers to the IT4LIA project in Italy and the MeluXina project in Luxembourg, both EU-backed. These are the reference deployments that turn a startup’s spec sheet into procurement-grade credibility. Getting designed into Dell and Supermicro systems at the same time — one via the enterprise channel, one via the hyperscaler-adjacent channel — gives Axelera two independent routes into customers who would never buy from a five-year-old Dutch company directly.

A crowded European field

Axelera is far from alone in attacking the inference market — chips that run AI models rather than train them. Its European peers and rivals include France’s VSORA, Britain’s Fractile, Spain’s Semidynamics, and the Netherlands’ own Euclyd. Notably, Euclyd announced a €200 million ($230.8 million) funding round on the same day, a reminder that capital is flooding into the continent’s silicon ambitions from multiple directions at once.

The competitive question is whether any of these players can reach the scale to matter. Axelera’s roadmap runs from Metis (edge) to Europa (enterprise) to Titania, a future chiplet-based architecture targeting data centres and supercomputers — Titania is backed by a €61.6 million EuroHPC Joint Undertaking grant. If the progression holds, Axelera becomes the first European startup with a credible end-to-end inference story. If it stalls at Europa, it risks becoming an acquisition target for a larger player — Samsung has already been reported as a backer, and the industry-wide consolidation of edge-AI silicon (Nvidia circling Rebellions, Qualcomm buying Edge Impulse) suggests buyers are circling.

Why it matters

For years the critique of European AI policy was that it produced papers and pilot projects while compute, models, and money stayed American. Axelera’s announcement is one of the first concrete data points against that critique: a European chip company with shipping product, named tier-one OEM channels, signed public-sector contracts, and a pipeline measured in billions. The $1.5 billion figure should be read sceptically — pipelines are aspirational by definition — but the tens of millions in signed deals are real, and the EU AI factory programme guarantees a baseline of demand that US rivals cannot easily displace.

The inference market itself is the fastest-growing slice of AI silicon, and the one where Nvidia’s grip is theoretically weakest. If Europa’s efficiency claims hold up in production, the interesting question is no longer whether Europe can build competitive AI chips, but whether it can build a competitive AI chip industry — and Axelera has just moved to the front of that queue.