Nvidia Bets $4B on US Optics
NVIDIA is investing $4 billion to de-risk its data center supply chain, announcing partnerships with Coherent and Lumentum. The company is putting $2 billion into each to expand U.S.-based manufacturing and R&D for next-gen optics, a critical component for AI infrastructure.
This investment is a direct response to the escalating power consumption and bandwidth density limitations of traditional copper interconnects in AI data centers. As AI models grow, the electrical signals in copper wires degrade over short distances, creating data bottlenecks and consuming significant power for signal boosting and cooling. The move specifically targets the advancement of co-packaged optics (CPO), a technology that integrates optical connectivity directly with silicon chips inside the same package. This proximity of optics to the processing units dramatically reduces the distance electrical signals need to travel, which in turn cuts down on power consumption and latency, two critical factors in large-scale AI clusters. Analysts project the CPO market will grow substantially, with some forecasts predicting it will exceed $20 billion by 2036, growing at a CAGR of 37% from 2026. This growth is driven by the necessity for faster and more efficient data transfer in AI infrastructure, with data rates of 800G and 1.6T becoming the new standard. This $4 billion investment is part of Nvidia's broader strategy of vertical integration, aiming to control the entire AI hardware stack, from GPUs to networking and interconnects. By securing the supply of critical optical components and driving R&D, Nvidia is not just a chip supplier but the architect of the entire AI data center, a move that mirrors Apple's own vertical integration with Apple Silicon. The partnerships with Coherent and Lumentum also aim to bolster US-based manufacturing and R&D in this critical sector, aligning with the objectives of the CHIPS and Science Act to strengthen the domestic semiconductor supply chain. The agreements are non-exclusive and include multi-billion-dollar purchase commitments and future access rights to advanced laser components and optical products. For AI workloads, the transition to optical interconnects is becoming essential. While copper is still effective for short-distance, scale-up connections within a server rack, optics are crucial for scale-out fabrics that connect thousands of GPUs across a data center. This hybrid approach is expected to define the architecture of next-generation AI systems.
