On March 2, 2026, NVIDIA made one of its boldest infrastructure moves yet: it announced two separate strategic agreements, investing $2 billion in Lumentum and another $2 billion in Coherent. Each deal also includes multibillion-dollar purchase commitments and future access or capacity rights for advanced laser and optical networking products. Read carefully, these are not ordinary supplier contracts. They amount to a declaration that the next bottleneck in AI is no longer computation alone, but the movement of data itself—and that light, not just electricity, will carry a growing share of that burden. (nvidianews.nvidia.com)
The logic behind this pivot has been visible for a year. At GTC on March 18, 2025, NVIDIA unveiled Spectrum-X Photonics and Quantum-X Photonics switches, describing them as a route to “million-GPU” AI factories. The company said these systems use co-packaged silicon photonics, integrating optics directly with switch silicon. In its materials, NVIDIA claimed 4x fewer lasers, 3.5x better power efficiency, far stronger signal integrity, and sharply improved resiliency compared with conventional pluggable optics. Its technical explanation is simple but profound: when signal paths shrink from inches to millimeters and external DSP overhead is reduced, power loss, latency, and failure points all fall. (nvidianews.nvidia.com)
That is why the $4 billion matters. NVIDIA is not merely buying components; it is helping finance the industrial base required to produce them. Lumentum said the investment will support R&D, future capacity, operations, and U.S.-based manufacturing through a new fab, while Coherent said NVIDIA’s funding will deepen R&D and expand U.S. manufacturing capacity for next-generation AI data centers. In other words, the company is moving upstream, into lasers, packaging, and fabrication readiness, because optical interconnects have become foundational to scaling AI infrastructure. (nvidianews.nvidia.com)
For learners of technology and language alike, this is the real lesson: “opticalization” is not a futuristic metaphor. It is an industrial strategy. NVIDIA’s Rubin-era roadmap already folds photonics into next-generation Ethernet networking, where the company says co-packaged optical systems can improve reliability, uptime, and performance per watt. The emerging contest in AI, then, is not simply who owns the fastest chips. It is who can build the least wasteful, most scalable fabric around them—and increasingly, that fabric is made of light. (nvidianews.nvidia.com)










