The technology landscape of Silicon Valley is undergoing a profound transformation driven by advances in networking technology—though not the kind involving social platforms like LinkedIn, but the sophisticated systems that connect computer chips and data centers. As the tech industry channels billions of dollars into building AI data centers, companies from chip manufacturing giants to emerging startups are innovating rapidly to improve the ways chips communicate with one another and with the infrastructure that supports them.
Networking has been a foundational element of computing since its earliest days. Initially, it connected large mainframe computers, enabling them to share data seamlessly. Today, networking in semiconductor technology operates at multiple levels within the computing stack. It involves the tiny interconnections between transistors on a chip, as well as the larger-scale connections that link servers and racks of servers in sprawling data centers. Efficient networking is essential to manage the enormous volumes of data that AI applications generate and process.
Established players like Nvidia, Broadcom, and Marvell have long been leaders in networking technology, supplying critical components that facilitate high-speed data transfer in data centers. However, the rapid growth and increasing complexity of AI workloads have spurred a new wave of innovation focused on accelerating data throughput beyond what traditional electronic interconnects can achieve. This is where deep-tech startups such as Lightmatter, Celestial AI, and PsiQuantum are making their mark by leveraging optical technology, or photonics, to push the boundaries of high-speed computing.
Photonics, which uses light to transmit information, has historically been considered an expensive and marginally useful technology. For over two decades, it languished on the sidelines of mainstream computing innovation. However, the AI boom has reignited interest in photonics as a promising solution to the bandwidth and speed limitations of conventional electronic networking. Pete Shadbolt, cofounder and chief scientific officer of PsiQuantum—a company specializing in optical quantum computing—describes photonics as having a “coming-of-age” moment fueled by the demands of AI.
Venture capitalists and institutional investors have recognized this potential and are investing billions into startups pioneering new photonics-based networking technologies. They see traditional electron-based interconnects as increasingly inadequate to handle the escalating data loads generated by AI applications, which require rapid, high-bandwidth communication between chips and data center components.
Ben Bajarin, a veteran technology analyst and CEO of Creative Strategies, notes that networking used to be a somewhat dull topic focused on moving packets of bits. But now, thanks to AI, networking must handle far more complex and intensive data workloads, prompting a surge of innovation aimed at increasing speed and efficiency. This shift has brought renewed attention to companies that can provide cutting-edge networking solutions.
Nvidia is often credited for its foresight in recognizing the importance of networking in AI infrastructure. In 2020, Nvidia made two strategic acquisitions that cemented its position in data center networking. The company acquired Mellanox Technologies for nearly $7 billion, a leader in high-speed networking solutions for servers and data centers. Shortly thereafter, Nvidia purchased Cumulus Networks, which develops Linux-based software for computer networking. These moves enabled Nvidia to enhance the performance of its GPUs by clustering them effectively in data centers, thereby maximizing their parallel computing power.
Beyond Nvidia, Broadcom stands out as a dominant force in custom chip accelerators and high-speed networking technology. The $1.7 trillion semiconductor giant collaborates closely with major tech companies like Google, Meta, and OpenAI to develop chips tailored for data center demands. Broadcom is also pioneering silicon photonics technology. According to a recent Reuters report, Broadcom is preparing to launch a new networking chip called Thor Ultra, designed to serve as a vital link between AI systems and the broader data center infrastructure.
ARM, a semiconductor design powerhouse, has also made strategic moves in the networking space. On its recent earnings call, ARM announced plans to acquire DreamBig, a company specializing in AI chiplets—small, modular circuit components that can be combined into larger chip systems. DreamBig’s technology is particularly important for scaling up and scaling out networking, meaning it helps connect components within a chip cluster as well as linking multiple racks of chips across a data center. ARM’s CEO Rene Haas emphasized the significance of DreamBig’s intellectual property for advancing networking capabilities.
Startups like Lightmatter are pushing the envelope with novel approaches to networking. Lightmatter’s CEO Nick Harris highlights the extraordinary pace at which AI computing power is growing—doubling roughly every three months, far surpassing the traditional pace described by