Nvidia invests $6B in optics for AI interconnects
Coherent, Lumentum, and Marvell receive capital to multiply InP wafer and silicon photonics production as copper reaches its limits in massive data centers.
June 20, 2026 · 3 min read
TL;DR: Nvidia has invested $6 billion in three optics companies (Coherent, Lumentum, Marvell) to expand production of optical components needed to interconnect thousands of GPUs in AI systems. The move aims to avoid supply chain bottlenecks and accelerate adoption of optical interconnects.
What happened?
In March 2026, Nvidia invested $2 billion in Coherent, a manufacturer of indium phosphide (InP) wafers for lasers and optical modulators. Coherent announced it will allocate $650 million to expand its plant in Sherman, Texas, quadrupling InP wafer production and creating 1,000 jobs. The investment is complemented by $20 million from the Texas Semiconductor Innovation Fund and up to $50 million from the CHIPS Act. Weeks later, Nvidia invested another $2 billion in Lumentum (transceivers, optical switches) and another $2 billion in Marvell (silicon photonics), totaling $6 billion. This investment offensive marks a strategic shift: instead of relying on purchase contracts, Nvidia is injecting capital directly into its critical suppliers to secure future production capacity. Coherent operates eight wafer fabs in the US, and the Sherman expansion will double its footprint, according to The Register.
Why is it important?
Next-generation AI systems, such as clusters of thousands of Blackwell or Rubin GPUs, require high-speed, low-latency interconnects that copper cannot provide over long distances. Integrated optics (silicon photonics, InP) enable links of hundreds of meters with superior energy efficiency. Jensen Huang, CEO of Nvidia, even said at Computex that Marvell could become the next trillion-dollar company thanks to optics. Without these technologies, horizontal scaling of AI would be limited. Historically, data centers have used copper for short distances and fiber optics for long links, but the density of AI clusters (thousands of interconnected accelerators) demands optics even within the rack. According to LightCounting, the data center optics market will grow from $12 billion in 2025 to over $30 billion in 2030. Nvidia's investment not only accelerates that growth but positions the company to control a critical bottleneck.
Consequences and context
The capital injection ensures the supply chain can meet expected demand starting in 2027. Coherent, Lumentum, and Marvell are key players in a market that, according to LightCounting, will grow from $12 billion in 2025 to over $30 billion in 2030. Coherent's expansion in Texas also boosts domestic semiconductor manufacturing in the US, aligned with the CHIPS Act. However, risks remain: demand may not materialize at the expected pace, and manufacturing InP and silicon photonics is complex with variable yields. Additionally, Nvidia is not the only customer; these manufacturers also serve competitors like AMD or Intel, which could dilute exclusive advantage. Nonetheless, Nvidia's investments include priority supply agreements, according to industry sources. Compared to past events, such as the $6.9 billion acquisition of Mellanox in 2020 to secure high-speed networking, this strategy is more decentralized: instead of acquiring companies, Nvidia takes minority stakes and funds expansions, maintaining supplier independence. It also echoes Apple's investment in Corning for iPhone glass, but on a much larger scale. The completion timeline for Coherent's expansion has not been disclosed; analysts expect it to take 2-3 years.
What readers should know
- Fiber optics and integrated photonics are essential for the next generation of AI data centers.
- Nvidia is securing its supply chain with direct investments, not just purchase contracts.
- InP technology is critical for laser sources; Marvell's silicon photonics targets massive integration.
- The Texas expansion will create jobs and strengthen US technological autonomy.
- The completion timeline for Coherent's expansion has not been disclosed; it is expected to take 2-3 years.
"Copper is no longer enough to scale AI to thousands of accelerators; optics is the new bottleneck and Nvidia is buying up all available capacity." — Analyst at TheVortiq.
In summary, Nvidia is playing the long game: with $6 billion, it is not only funding capacity but sending a signal to the market that optics will be as crucial as GPUs. Competitors will need to follow suit or risk falling behind in the AI infrastructure race.