Vsora pitches itself as Europe’s Nvidia for inference

Vsora CEO Khaled Maalej positioned the French chipmaker as a European alternative to Nvidia and AMD for AI inference, arguing that its flagship Jotunn8 accelerator benefits from rare access to HBM3E memory and TSMC’s CoWoS advanced packaging. The technology places high-bandwidth memory alongside the AI accelerator die in the same package, a design intended to address the memory wall that limits efficient inference.

Vsora has shifted from digital signal processing IP into selling its own silicon, building on Maalej’s earlier work with Dibcom and long-running ties to Taiwan’s semiconductor ecosystem. Jotunn8 is described as offering 288 gigabytes of memory and 3.2 petaflop processing power, with the company aiming to reduce infrastructure cost and energy use for AI workloads.

The company remains focused on silicon rather than selling complete systems, while working with partners on servers and racks to accelerate deployment. Vsora has around 70 staff across Asia and the Middle East, plans a next-generation chip later this year, and is expanding toward the U.S. after a recent fundraising round.

Maalej tied the company’s pitch to growing demand for sovereign compute, saying data center operators want at least 10-20% of chips from outside the U.S. He argued that Europe needs its own AI silicon as compute becomes increasingly fundamental.