Nvidia invests $2B each in Coherent and Lumentum to secure silicon photonics supply for AI datacenters, highlighting the critical role of optical networking in next-generation AI infrastructure.
Nvidia is making a massive $4 billion bet on American photonics manufacturing, investing $2 billion each in Coherent and Lumentum to secure supply of critical optical components for its AI datacenter infrastructure.
The investment, announced alongside multibillion-dollar purchase commitments, underscores how silicon photonics has become essential to Nvidia's networking business, which generated over $31 billion in revenue during its 2026 fiscal year. Both companies specialize in optical components like pluggable transceivers, laser sources, and optical circuit switches used in datacenter optics.
Why Silicon Photonics Matters for AI
While Nvidia has resisted photonics for its scale-up NVSwitch fabrics due to power concerns - sticking with copper saves approximately 20kW in its 120kW GB200 NVL72 systems - the company has embraced silicon photonics for scale-out networks. These networks connect multiple systems or racks for distributed AI inference and training, where power efficiency becomes critical.
In 2025, Nvidia unveiled its next-generation Spectrum and Quantum switches would use co-packaged optics, integrating optical transceivers directly into the switch. This approach dramatically reduces the number of optics required by eliminating pluggable modules on the switch side, cutting power consumption significantly.
For now, Nvidia maintains pluggable optics on the NIC side, but the integrated switch design still relies on laser modules as light sources. The company also needs optics for products like Spectrum-XGS, which distribute workloads across multiple datacenters at ranges starting at 500 meters.
Manufacturing Expansion in the US
As part of this investment, both Coherent and Lumentum plan to expand their manufacturing capacity in the United States. This move aligns with broader efforts to reshore critical technology manufacturing and reduce supply chain dependencies.
The timing is notable, coming just days after OpenAI announced a $110 billion funding round in which Nvidia would contribute $30 billion. This demonstrates how deeply intertwined Nvidia's hardware and software ecosystem has become with the AI industry's growth.
The Future of Photonics in AI Infrastructure
Many companies in the space, including Ayar Labs and Lightmatter, believe that as bandwidth demands continue to increase, chip designers like Nvidia will eventually be forced to embrace silicon photonics interconnects for scale-up networking as well. Such a design would see fiber optic cables connected directly to each accelerator, potentially revolutionizing how AI systems communicate internally.
However, not everyone shares this optimism. Broadcom's CEO has stated that silicon photonics won't matter "anytime soon," highlighting the ongoing debate about when and how these technologies will mature.
Strategic Implications
This investment represents more than just securing supply - it's a strategic move to ensure Nvidia maintains its competitive edge in AI networking. With competitors like Cisco challenging Broadcom and Nvidia with high-speed switches, controlling the photonics supply chain becomes increasingly important.
The deal also reflects the broader trend of AI infrastructure becoming a critical bottleneck in the industry's growth. As AI models continue to scale, the networking fabric that connects thousands of GPUs becomes as important as the GPUs themselves.
For datacenter operators and AI companies, this investment signals that silicon photonics is no longer a niche technology but a core component of next-generation AI infrastructure. The billions flowing into this space suggest we're at the beginning of a significant shift in how datacenters are built and operated.
Comments
Please log in or register to join the discussion