The rapid growth of artificial intelligence (AI) is transforming the tech industry, and at its heart lies a crucial element often overlooked: the technology that connects computer chips. As companies pour billions into building massive AI data centers, chip makers are racing to develop faster, more efficient ways for chips to communicate with each other and with external systems.
This isn’t just about upgrading existing networking infrastructure; it’s about creating entirely new approaches to handle the deluge of data powering advanced AI applications.
Traditionally, semiconductor networking has existed at multiple levels: from interconnecting transistors within a single chip to linking server racks filled with powerful processors. Established players like Nvidia, Broadcom, and Marvell already have strong footholds in this market. However, the demands of AI are pushing the boundaries of what’s possible, prompting innovative startups to explore unconventional solutions.
Enter optical technology, or photonics – a field that has seen renewed interest thanks to the AI boom. While once considered too expensive and complex, photonics uses light to transmit data instead of electrons, offering the potential for significantly faster speeds and lower power consumption. Companies like Lightmatter, Celestial AI, and PsiQuantum are leading this charge, developing advanced silicon photonic systems designed specifically for high-performance computing demands.
Lightmatter CEO Nick Harris highlights a key challenge: “The amount of computing power required for AI now doubles every three months—far faster than Moore’s Law predicts.” As chips grow in size and complexity, efficient interconnectivity becomes paramount. Their solution involves stacking silicon chips together and connecting them via light-based interconnects, essentially creating ultra-fast 3D chip clusters.
Other startups are also making waves. Celestial AI recently secured $250 million in funding from prominent investors including AMD and Fidelity Management, while PsiQuantum, focused on building quantum computers using photonics, raised $1 billion led by BlackRock. These investments reflect the growing confidence that optical networking holds the key to unlocking future computing capabilities.
However, there are hurdles to overcome. Photonics is still relatively expensive to manufacture and integrate with existing electrical systems. Additionally, established giants like Broadcom and Marvell have deep expertise in serving hyperscalers (companies like Google and Meta) with both AI chips and customized networking solutions. Their ability to scale rapidly and cater to specific client needs gives them a significant advantage.
While the photonics revolution might still be some years away, the race is on. The increasing demand for faster data speeds in AI applications ensures that innovative networking technologies will continue to attract investment and drive competition within this vital sector of the tech industry. As Ben Bajarin, a longtime tech analyst at Creative Strategies, aptly puts it: “Networking is the thing that makes computers function, but it just feels like the industry is moving towards much more customization.” This shift could create opportunities for nimble startups, even as established players maintain their dominance.
