Since March 2026, NVIDIA has invested at least $6.5 billion in photonics companies. This is the largest bet any single company has placed on photonics in the industry's history, but the company is not buying these businesses outright. It is taking equity stakes, signing multi-year supply agreements, and funding new factories. The goal is making sure that advancements in photonics continue and it will prevent them from hitting a scalability and performance wall that will occur if they remain on electrical and copper.
Nvidia CEO Jensen Huang said at GTC in March (figure 1), pointing to Nvidia’s ethernet networking platform used to connect AI factories and GPU clusters.
“When you look upstream, you come to the conclusion that we’re starting to scale our silicon photonics technology, which means the amount of silicon photonics technology capacity that we need is substantially higher than the world has today. So we work with the supply chain to make sure we can help them build up the capacity in advance of that.”
Figure 1. Nvidia CEO Jensen Huang giving a keynote speech.
The problem NVIDIA is trying to solve
Modern AI systems need thousands of GPUs working together as one machine, and most of the wiring between them is still copper. Copper is cheap and reliable, but it wastes more energy as data speeds rise. At the scale NVIDIA is building now, that waste becomes a real constraint, limiting how many GPUs fit in a rack and how much power is left over for actual computing.
Optical connections avoid this problem. Light-based data links carry information over longer distances at much lower power. That is why NVIDIA is pushing an approach called co-packaged optics, where optical components sit right next to the processor instead of on a separate card.
The Deals
NVIDIA's spending breaks down into five main moves.
In March, NVIDIA invested $2 billion each in Coherent and Lumentum, two established makers of lasers and optical components. Both deals came with large, multi-year purchase agreements and both companies are expanding US manufacturing with the money.
“This strategic relationship underscores Coherent’s role as a key enabler of next-generation AI data center infrastructure. We are proud to expand our 20-year relationship with NVIDIA by increasing their access to include multiple product families to help them build the AI data centers of the future.”
“This multiyear strategic agreement reflects our shared commitment to advancing the optics technologies that will power the next generation of AI infrastructure. In support of this collaboration, we are also investing in a new fabrication facility to increase capacity and accelerate innovation. We’re excited to work together to expand what’s possible for the AI optical architectures of tomorrow.”
A third $2 billion investment went to Marvell, a chipmaker that had already made its own move into photonics. In February, Marvell closed a separate $3.25 billion acquisition of Celestial AI, a startup building a Photonic Fabric that routes optical data directly into a processor rather than just to its edge. NVIDIA's investment in Marvell sits alongside that acquisition rather than being part of it, but both are telling that optical interconnects are becoming central to how AI chips communicate.
In May, NVIDIA on a multi-year partnership with Corning, the glassmaker known for inventing low-loss optical fiber. The deal opened with a $500 million equity stake and gives NVIDIA the right to invest further through stock warrants. Corning CEO Wendell Weeks called the partnership “nothing short of extraordinary, not just for the future of AI.” Corning is using the funding to build three new manufacturing plants in North Carolina and Texas, a move expected to expand its US optical connectivity capacity tenfold.
The fifth deal was smaller and less direct. NVIDIA joined AMD, MediaTek, and a group of institutional investors in a $500 million funding round for Ayar Labs, a startup building optical chiplets that connect directly to GPUs and switches. The round valued Ayar Labs at $3.75 billion. Ayar Labs CEO Mark Wade said the technology was designed for “thousands of GPUs to operate as a unified system.”
What it means commercially
Two things stand out about how NVIDIA structured these deals. First, nearly every one came bundled with a multi-year purchase commitment. NVIDIA is locking in supply years in advance, at a moment when every other AI chip maker wants the same components. The scale of that lock-up shows up in Lumentum's own order book. Hurlston said in March: “We're sold out really until the end of 2027. We see no end in sight.” Trade press has reported that NVIDIA's purchase agreements with Coherent and Lumentum could tie up a large share of the world's high-end laser supply through 2027.
Second, competitors are pooling resources rather than going it alone. AMD and MediaTek joined NVIDIA in the Ayar Labs round instead of each backing a separate rival, and AMD has separately acquired the startup Enosemi and taken stakes in Teramount and Celestial AI. Hyperscalers are moving too: Alphabet and Microsoft backed the optical startup nEye in an $80 million round in April. That spread of activity shows how tight the supply base for advanced photonics remains. Only a handful of companies can produce these components at the volumes AI infrastructure now requires, and none can scale fast enough alone.
The market has taken notice. As of late May, Lumentum's share price had risen 134 percent since the start of the year, with Coherent up 96 percent, Marvell up 122 percent, and Corning up 111 percent, according to CNBC. Photonics has become one of the more visible trades tied to the AI infrastructure buildout, and investors are pricing these deals as validation of the whole sector, not just of NVIDIA's strategy.
The investments also carry a domestic manufacturing angle: every major deal funds new US production. NVIDIA has tied this to competition with Chinese AI hardware, treating a secure domestic supply chain as a strategic asset, not just a cost decision. For an industry where key materials such as indium phosphide are concentrated in a small number of countries, that reshoring push has consequences well beyond NVIDIA's own supply chain.
Not without risk
Money does not remove the engineering problem. “The technology is sound, production scale is the harder problem,” said Nick Patience, AI lead at the Futurum Group, in comments to CNBC. Patience pointed to manufacturing yield as the real obstacle: aligning optical and electronic components with the precision AI hardware demands leaves little room for error, and a single misaligned part can force a rework. NVIDIA's billions can build factories and lock in supply, but they cannot shortcut the slow work of getting yields up to hyperscale volumes.
The Bigger Picture
None of this guarantees that co-packaged optics, or any single architecture, wins outright. Broadcom is pursuing its own approach with the Tomahawk 6 switch, and Celestial AI's technology differs from the pluggable optics still used across most of the industry today. What NVIDIA's spending does confirm is that the bottleneck has moved. For years, the AI industry worried about chip supply. Now the constraint is shifting to the optical components that connect those chips together, and the company best positioned to profit from AI compute is also the company writing checks to make sure that constraint does not slow it down.