Kandou AI raises $225 million to extend copper interconnects in AI infrastructure

Kandou AI, a Swiss developer of chip-to-chip communication technology, has raised $225 million in a round the company calls a Series A. The startup, valued at $400 million, is betting that copper interconnects have not yet exhausted their potential and will remain an important part of AI infrastructure.

Why the round stands out

This is not just another investment in hardware. Kandou works in a layer that rarely makes headlines but directly affects the performance of AI systems: moving data between chips. As models grow and accelerators are packed more densely into servers and racks, the bottleneck is increasingly not computation itself but the speed and efficiency of data exchange. That is why money is flowing not only to the makers of the chips, but also to companies solving the problem of connecting them.

The amount stands out too. For a company of this profile, $225 million signals that investors are willing to fund not only high-profile platforms around generative AI, but also the basic infrastructure without which data center scaling runs into the limits of cost, power consumption and assembly complexity. The $400 million valuation shows that the market sees Kandou not as a niche component supplier but as a potentially important player in the AI hardware supply chain. It is another reminder that the most expensive places in the AI stack now sit deep beneath the application layer.

A bet on copper

Kandou's central idea is clear from the headline of the deal: the company is convinced that copper connections can last longer than advocates of a rapid shift to optics expect. Amid the AI infrastructure boom, optical interconnect solutions are often seen as the inevitable future, especially where large volumes of data must move with minimal loss. But inside servers, accelerator modules and short links, copper retains strengths, from price to manufacturing maturity and compatibility with existing architectures.

Kandou builds chip-to-chip interconnect technology, meaning it works at the level where bandwidth, latency, power consumption and packaging density matter. If the company can improve these parameters on copper lines, it extends the economic life of existing approaches and eases the pressure on data center operators, who do not have to rebuild all their physical infrastructure for optics overnight. In other words, the bet is not against light as such, but for a more pragmatic transition path.

Who backed the round

The list of participants shows that this goes beyond a classic venture round. Maverick Silicon led the round, and SoftBank, Synopsys, Cadence Design Systems and Alchip Technologies also took part. That matters because alongside financial capital, players tied to the design, supply and scaling of semiconductor solutions are coming into the project. An investor list like this often means more than money: it can also mean access to industry connections, customers and future integrations.

• Maverick Silicon — lead investor of the round
• SoftBank — strategic participant with a large bet on the AI ecosystem
• Synopsys — a key player in chip design tools
• Cadence Design Systems — another important supplier of EDA tools
• Alchip Technologies — a partner involved in bringing specialized chips to market

The Series A label also draws attention. With a round of this size, this set of participants and a $400 million valuation, the company looks more mature than a typical early-stage startup. That is not necessarily a contradiction: in infrastructure semiconductor business, cycles are longer, and the round label sometimes reflects a company's maturity worse than its list of partners and the amount of capital raised. In Kandou's case, what matters is not the letter on the slide but who put money behind this thesis.

What it means

Investors keep looking not only for winners among models and clouds, but also for those solving the physical constraints of AI infrastructure. If Kandou can prove that copper interconnects keep their advantage in price and efficiency over short distances, the move to fully optical systems may turn out to be a drawn-out evolution rather than a revolution. For the market, that means a longer life for hybrid architectures and renewed interest in companies improving the base layer of data centers.