Nvidia is pouring billions of dollars into photonics technology as the artificial intelligence boom exposes a growing problem at the heart of modern AI infrastructure: moving vast amounts of data fast enough without overwhelming power grids and data centers.
Over the past three months alone, Nvidia has committed at least $6.5 billion to companies developing optical and silicon photonics technologies, marking one of the clearest signs yet that the AI race is shifting beyond graphics processors and into the networks that connect them.
The investments span a wide range of companies tied to optical connectivity. Nvidia committed $2 billion in investments into Lumentum, Coherent and Marvell Technology. It also pledged $500 million to Corning for advanced optical connectivity development and joined a $500 million funding round for startup Ayar Labs.
The scale of the spending underlines how rapidly the AI industry’s bottlenecks are evolving. For years, the focus centered almost entirely on access to GPUs. But as AI models become larger and more computationally intensive, the challenge increasingly lies in moving information efficiently between processors, servers, and entire AI clusters.
Photonics, which uses light instead of electrical signals to transmit data, is emerging as one of the industry’s most promising solutions. Existing copper-based systems consume substantial amounts of electricity and generate heat as AI workloads intensify. Analysts say that could become a major constraint on the expansion of AI infrastructure globally.
Alvin Nguyen, senior analyst at Forrester, said Nvidia’s investment strategy reflects growing concern that traditional electrical interconnects may not scale alongside AI demand.
“Photonics represents a way for Nvidia to scale their AI infrastructure without the energy costs that staying with electrical and copper will incur,” Nguyen told CNBC.
The issue has become particularly urgent as hyperscalers and AI developers build massive GPU clusters. Future AI systems are expected to require millions of interconnected chips operating simultaneously across multiple data centers. That scale creates enormous networking demands.
Nvidia Chief Executive Jensen Huang has repeatedly warned that existing infrastructure will struggle to keep pace with the next generation of AI factories.
At Nvidia’s GTC conference in March, Huang said the company was already integrating photonics into networking systems and GPU-to-GPU interconnect technology. He added that future AI deployments would require far more silicon photonics manufacturing capacity than currently exists worldwide.
Morningstar analyst Brian Colello said Nvidia’s next-generation rack-scale AI systems will require exponentially greater bandwidth as models become more advanced and AI usage expands globally.
“Nvidia’s roadmap of next generation AI rack-scale solutions will require an increasing amount of optical connectivity,” Colello told CNBC.
The investment surge has also fueled a dramatic rally in photonics-related stocks. Shares of Lumentum have climbed 134% this year, while Coherent has gained 96%. Marvell has risen 122%, and Corning more than doubled as investors increasingly view optical networking as a critical pillar of the AI economy.
Nvidia is not alone in chasing photonics technologies. Advanced Micro Devices (AMD) has also invested in Ayar Labs and acquired startup Enosemi in 2025, while making strategic bets on companies including Teramount and Celestial AI. Venture arms tied to Alphabet and Microsoft backed startup nEye earlier this year.
The growing investor interest suggests the industry views optical infrastructure as essential to sustaining AI growth through the end of the decade.
Still, analysts caution that photonics deployment at scale remains technically difficult.
Nick Patience, AI lead at The Futurum Group, said manufacturing complex optical assemblies remains one of the industry’s toughest engineering challenges because even small alignment errors between optical and silicon components can render systems unusable.
“The technology is sound, production scale is the harder problem,” Patience said.
That means widespread adoption across the AI infrastructure stack may still take years. Analysts expect large-scale deployment to accelerate closer to 2028 as manufacturing processes mature and costs decline.
Yet Nvidia’s aggressive investment pace suggests the company sees little room for delay. The AI boom has already strained electricity supplies, data center construction pipelines, and semiconductor manufacturing capacity worldwide. If the compute infrastructure cannot move data efficiently enough, the performance gains from more advanced AI chips risk being bottlenecked by the network itself.
