Kandou AI Secures $225M to Tackle AI Memory Wall

In a significant move to address a critical hardware bottleneck, Swiss semiconductor firm Kandou AI has secured a $225 million investment. This funding round, labeled a Series A, was led by Maverick Silicon and included major strategic players like SoftBank, Synopsys, Cadence Design Systems, and Alchip Technologies. The deal values the company at $400 million. While the “Series A” moniker suggests a new venture, Kandou has a long history, having been founded in 2011 as Kandou Bus and raising over $163 million in prior rounds. The new designation reflects a strategic pivot under fresh leadership, not a startup launch.

The company’s new direction is spearheaded by CEO Srujan Linga, a former Goldman Sachs managing director who took the helm in 2025. He succeeded founder Amin Shokrollahi, a professor whose foundational invention remains central to the business. That core technology is a signalling method called Chord signalling, which transmits correlated signals across multiple wires. This approach can double or quadruple bandwidth while cutting power consumption in half. Linga’s rebranding of the company to Kandou AI and its sharpened focus on AI infrastructure has clearly resonated, attracting the largest investment in its history and bringing SoftBank, a prolific investor in the AI space, into the fold.

Kandou’s strategy is notable for its contrarian material choice in solving a widely acknowledged problem. The AI memory wall and interconnect bottleneck are severe constraints as models grow and GPU clusters expand. Data movement between processors and memory now limits overall performance. At high speeds, traditional copper interconnects consume about 30 percent of a cluster’s power and suffer from signal degradation, restricting their effective range to under a meter.

The industry’s dominant answer has been a shift to optical interconnects. Companies like Ayar Labs, which recently raised $500 million, and Celestial AI, acquired by Marvell for $3.25 billion, are advancing photonic technologies that promise vastly superior bandwidth and latency. The market for optical solutions in AI data centers is forecast to grow dramatically, from $3.75 billion in 2025 to over $18 billion by 2033.

Kandou AI, however, is making a bold bet that copper interconnects still have a future. The company asserts its Chord technology can achieve path-to-Shannon-capacity efficiency, potentially reducing power and system costs by a factor of ten. It aims to extend copper link speeds to 448 gigabits per second and beyond. If successful, this would challenge the assumption that a wholesale shift to optics is immediately necessary, suggesting existing copper infrastructure could be upgraded cost-effectively for several more hardware generations.

The roster of strategic investors reveals the intended path to market. Participation from Synopsys and Cadence, the leaders in electronic design automation (EDA) tools, suggests plans to integrate Kandou’s serializer/deserializer IP directly into industry-standard chip design workflows. Alchip provides a bridge to manufacturing, while SoftBank contributes both substantial capital and a vast strategic network in AI.

This points to a licensing and IP model akin to Arm’s approach, where Kandou’s technology would be embedded in chips designed by other companies. This capital-efficient strategy avoids the immense costs of manufacturing and selling proprietary silicon. The central challenge for Kandou is to execute this model against well-funded optical rivals, despite its comparatively modest $400 million valuation.

That valuation gap is stark. Kandou is worth roughly a tenth of Ayar Labs and a fraction of the Celestial AI acquisition price. This may reflect skepticism about copper’s long-term viability, or it could indicate a market underestimating the advantage of leveraging existing infrastructure. Copper is already ubiquitous in data centers. If Kandou’s technology can make it performant for the next wave of AI workloads, adoption could be faster and cheaper than a full optical overhaul.

The primary risk is that the window of opportunity may be short. AI model scaling continues to outpace infrastructure roadmaps. Speeds that seem sufficient today at 448 gigabits per second may be obsolete when next-generation models demand terabit-per-second transfers in just a few years. Optical interconnects, despite their cost and complexity, offer a higher theoretical performance ceiling.

This $225 million infusion gives Kandou AI time to prove its thesis. The company’s 15-year history and the proven commercial deployment of Chord signalling in consumer electronics lend technical credibility. Yet the AI infrastructure market often rewards revolutionary ambition over pragmatic incrementalism. Advocating for an upgrade to the existing material is a more difficult narrative than promising to replace it entirely. The investors in this round are backing engineering pragmatism. Whether the broader market agrees will hinge on the accelerating timeline of the optical transition and on whether enhanced copper can keep pace with an industry that refuses to wait.