As AI models grow, the need for ultra‑high‑bandwidth data movement between chips is rapidly becoming one of the biggest infrastructure challenges in modern computing. Traditional interconnect technologies were not designed for AI‑scale workloads, and bandwidth limitations are increasingly capping performance across hyperscale data centres.
Based in Saint‑Sulpice, Switzerland, Kandou, a spin‑off from École Polytechnique Fédérale de Lausanne (EPFL), is tackling that problem through high‑speed copper‑based interconnect technology. Its chips are engineered to move data faster and more efficiently between GPUs, CPUs, and memory modules, using advanced signal‑integrity and equalisation techniques that reduce latency and power consumption.
According to industry reports, the company has raised around $225 million in a Series A round to accelerate its work in this area. The round was said to include a mix of financial and strategic backers, reportedly including SoftBank Group, Synopsys, Cadence Design Systems, and Alchip Technologies.
Founded by Dr Amin Shokrollahi, a mathematician and EPFL professor whose coding theory research underpins Kandou’s signal‑transmission methods, the company has established itself as one of Europe’s most prominent innovators in low‑power, high‑bandwidth chip‑to‑chip links.
Kandou’s copper‑based interconnect platform, marketed under its Chord™ signalling architecture, has already been adopted in high‑volume applications, with over 20 million silicon units shipped worldwide.
Direct competitors include Ayar Labs, Lightmatter, Celestial AI, and ZeroPoint Technologies.
The new capital, if confirmed, will support scaling production of AI‑centric connectivity chips, expanding global partnerships with hyperscalers, and developing multi‑terabit interconnect solutions optimised for next‑generation AI and cloud systems.
