NVIDIA Invests $4 Billion in Lumentum and Coherent as Photonics Emerges as AI's Next Supply Chain Bottleneck
NVIDIA pours $2 billion each into photonics makers Lumentum and Coherent, signaling that optical interconnects are becoming as critical to AI scaling as GPUs.
Overview
NVIDIA announced on March 2, 2026, that it will invest $2 billion each in photonics companies Lumentum Holdings and Coherent Corp., pairing the capital infusions with multibillion-dollar, multiyear purchase commitments for advanced laser and optical networking components. The $4 billion combined outlay marks the chipmaker’s most aggressive move yet to secure the optical supply chain that underpins its next-generation AI data center architecture.
What We Know
The two deals follow an identical template. According to NVIDIA’s press release on the Lumentum partnership, each agreement is nonexclusive and bundles a $2 billion investment with guaranteed future purchase orders and capacity access rights for advanced laser components and optical networking products. Both Lumentum and Coherent have committed to expanding U.S.-based manufacturing, with NVIDIA’s Coherent announcement noting the investment will support research and development, future capacity, and operations as Coherent builds out new domestic fabrication capabilities.
“Together with Lumentum, NVIDIA is advancing the world’s most sophisticated silicon photonics to build the next generation of gigawatt-scale AI factories,” NVIDIA CEO Jensen Huang said in the Lumentum press release. In a parallel statement for the Coherent partnership, Huang described the transition as inevitable: “Computing has fundamentally changed…NVIDIA is pioneering next-generation silicon photonics to enable AI infrastructure at unprecedented scale, speed and energy efficiency.”
The two companies occupy complementary positions in the optical stack. Coherent specializes in co-packaged optics integration, fiber-to-chip connectors, and optical packaging, while Lumentum focuses on high-power continuous-wave laser chips used as external light sources. The dual-vendor structure builds depth across distinct optical layers rather than simple supplier redundancy.
NVIDIA already uses co-packaged optics in its Spectrum-X and Quantum-X networking switches, integrating optical transceivers directly into switch packages. According to The Register, the company’s networking business generated over $31 billion in revenue during fiscal 2026, underscoring the financial scale at which optical components now operate.
Why Photonics Has Become Critical
The investments address a fundamental physical constraint. As AI clusters scale to tens of thousands of GPUs, traditional copper interconnects are reaching practical limits at the 800 Gbps data rates that modern workloads demand. Optical systems transfer data through light via glass, delivering superior energy efficiency and lower latency. NVIDIA claims its co-packaged silicon photonics deliver 3.5 times lower power consumption and up to 10 times better network resiliency compared to traditional pluggable optical transceivers.
The bottleneck has a supply-side dimension as well. Indium phosphide laser fabrication capacity, essential for producing the laser sources that drive optical interconnects, is severely limited. Current transceiver demand reportedly exceeds supply by a factor of two due to these constraints, mirroring the advanced packaging shortage that constrained GPU supply in 2023 and 2024. With single AI racks now drawing up to 600 kilowatts and optical networking consuming roughly 10 percent of that power envelope, the efficiency gains from photonics are no longer optional.
As CNBC reported, the deals represent NVIDIA’s strategy to shore up research pipelines and supply chains to support the major AI infrastructure buildout underway across hyperscale data centers.
What We Don’t Know
Neither NVIDIA nor its partners have disclosed specific timelines for when the expanded manufacturing capacity will come online, though analysts estimate indium phosphide laser fabrication could take 18 to 24 months to ramp. The nonexclusive nature of both agreements leaves open the question of whether hyperscalers such as Amazon and Google, which have pursued their own silicon photonics strategies, will also secure capacity from Lumentum and Coherent.
The deals also do not address whether NVIDIA will eventually adopt photonics for its scale-up NVSwitch fabric, which connects GPUs within a single node. According to The Register, the company has so far resisted photonics for scale-up interconnects due to power consumption considerations, limiting optical links to scale-out rack-to-rack networking.
China’s concentration of indium supply introduces a geopolitical variable. Should trade restrictions tighten further, the emphasis on domestic U.S. manufacturing in both deals could prove strategically significant beyond pure economics.
Analysis
The $4 billion investment follows a playbook NVIDIA has used before. When high-bandwidth memory and advanced packaging threatened to constrain GPU shipments, the company moved upstream with direct investments and guaranteed purchase commitments to compress multi-year supply shortfalls. The photonics deals apply the same logic to a different chokepoint: the optical interconnects that stitch together the massive GPU clusters powering AI training and inference.
With NVIDIA’s networking revenue already exceeding $31 billion annually, photonics has moved from a peripheral concern to a core strategic priority. The market’s response reflected that shift, with Lumentum shares rising nearly 12 percent and Coherent jumping 15 percent on the announcement day.