Nvidia’s $4 Billion Optics Bet: How Coherent and Lumentum Became AI Infrastructure Gatekeepers
Chip giant locks in optical component supply with strategic investments as data center interconnect market sprints toward $31 billion by 2033.
Nvidia announced $2 billion investments in both Coherent and Lumentum on Monday, securing multi-billion dollar supply commitments for optical components as the race to build gigawatt-scale AI data centers intensifies competition for specialized photonics manufacturing capacity. The deals, structured as non-exclusive partnerships, pair equity stakes with guaranteed purchase volumes and priority access to advanced laser components and optical networking products. Nvidia will invest $2 billion in Lumentum and the same amount in Coherent, with both agreements including multi-billion dollar purchase commitments and future access and capacity rights for advanced laser and optical networking products. Shares of Lumentum and Coherent rose more than 7% in premarket trading after the announcement.
The capital deployment reflects a fundamental architectural shift inside AI factories. Optical interconnect technology and package integration are critical for the continued scaling of AI factories, improving the energy efficiency and resiliency of large-scale AI networks. Jensen Huang, Nvidia’s CEO, stated that CNBC reported the partnerships will advance ‘the world’s most sophisticated silicon photonics to build the next generation of gigawatt-scale AI factories.’
The $20 Billion Opportunity: Why Copper Can’t Keep Up
Traditional electrical interconnects are reaching physical limits as AI clusters scale beyond 100,000 GPUs. GPU to GPU or switch links require much higher bandwidth and lower latency to enable GPUs to be interconnected so that they can coherently share resources such as memory. 5th generation NVLink on Nvidia Blackwell offers 900GByte/s (7,200 Gbit/s) of uni-directional bandwidth per GPU. That is 9x more bandwidth per GPU than the 100GByte/s (800Gbit/s) per GPU on the back-end scale out network, according to SemiAnalysis.
The market is responding. The optical interconnect in AI Data Centers market reached $9.94 billion in 2025 and is expected to reach $31.04 billion by 2033, growing with a CAGR of 15.3% during the forecast period 2026-2033, according to DataM Intelligence. The more aggressive co-packaged optics segment is forecast to grow even faster: IDTechEx forecasts that the co-packaged optics market will rise at a robust compound annual growth rate (CAGR) of 37% from 2026 to more than $20 billion by 2036, reported by Semiconductor Today.
- 3.5x better power efficiency in co-packaged optics architectures
- 10x improved network resiliency by reducing component count
- 63x greater signal integrity at 1.6 Tb/s port speeds
- Enables scale-up domains beyond the 2-meter copper limit
Manufacturing Fortress: U.S. Fab Expansion and Supply Chain Lock-In
Both investments fund U.S.-based capacity expansion. Nvidia is investing $2 billion in Lumentum to support R&D, future capacity and operations as the company builds out its U.S.-based manufacturing capabilities in a new fab. Nvidia is investing $2 billion in Coherent to support research and development, future capacity and operations as Coherent builds out its U.S.-based manufacturing capabilities, according to statements from Nvidia’s official newsroom.
The timing aligns with geopolitical Supply Chain pressures. Coherent’s recent earnings highlight the demand surge: Revenue grew 17% year-over-year to $1.69 billion. Coherent’s Q2 2026 results reflect a robust performance, driven by a 17% year-over-year increase in revenue, according to Investing.com. CEO Jim Anderson stated that management expects fiscal 2027 revenue growth to exceed fiscal 2026’s pace, driven by AI data center buildouts.
Lumentum’s manufacturing roadmap now includes dedicated capacity for 800G and 1.6T transceiver components. LightCounting is forecasting sales of 1.6T Ethernet optical transceivers to begin ramping in 2026, and to exceed $15 billion by 2030, reported by Acacia.
| Source | 2025 Value | 2030+ Target | CAGR |
|---|---|---|---|
| DataM Intelligence (AI Data Centers) | $9.94B | $31.04B (2033) | 15.3% |
| IDTechEx (Co-Packaged Optics) | ~$1.0B | $20B+ (2036) | 37% |
| LightCounting (1.6T Transceivers) | Ramp start | $15B (2030) | – |
| Mordor Intelligence (Total Optical) | $21.88B | $40.03B (2031) | – |
Silicon Photonics: The Packaging War Heats Up
Nvidia’s strategy extends beyond pluggable transceivers. In March 2025, the company unveiled Spectrum-X and Quantum-X photonics switches using co-packaged optics, where optics innovations with 4x fewer lasers deliver 3.5x more power efficiency, 63x greater signal integrity, 10x better network resiliency at scale and 1.3x faster deployment compared with traditional methods, according to Nvidia Investor Relations.
TSMC’s COUPE (Compact Universal Photonic Engine) process has become the foundry standard for integrating optical chiplets alongside GPUs and memory. All silicon photonics chiplets that TSMC integrates will be made by TSMC. For some time, TSMC has been developing its COUPE process, explained in Semiconductor Engineering. Both Nvidia and Broadcom have begun deploying COUPE in scale-out switches, consolidating photonic manufacturing around the world’s largest chip foundry.
Co-packaged optics (CPO) integrates optical transceivers directly with switch ASICs on a common substrate, eliminating pluggable modules. This reduces the electrical path from centimeters to micrometers, cutting power consumption by 50% and enabling ports beyond 1.6 Tb/s. However, CPO introduces serviceability risks—if the optical engine fails, the entire switch may be unusable, unlike pluggable transceivers that technicians can swap in minutes.
Competitive Dynamics: Who Benefits, Who Loses
The investments validate the optical supply chain but create winners and losers. Fabrinet, the contract manufacturer assembling modules for both Coherent and Lumentum, stands to gain from volume ramps. Broadcom, the dominant supplier of switch ASICs and photonic DSPs, competes with Nvidia on CPO but also supplies components to both optical vendors.
Chinese suppliers face pressure. Potential new tariffs on Chinese imports, which could be reintroduced under a Trump administration, stand to benefit domestic optical component suppliers like Lumentum, Coherent, and Applied Optoelectronics, noted Industrial Analyst. Innolight, a Shenzhen-based transceiver vendor with significant hyperscaler share, could see U.S. market access restricted if geopolitical tensions escalate.
The deals also signal hyperscaler unbundling strategies. Historically, cloud providers bought complete GPU systems from Nvidia, including optical transceivers. Some hyperscalers are rumored to be taking an unbundling approach. This refers to hyperscalers purchasing GPUs from Nvidia, but sourcing optical components directly from suppliers to lower cost and create customized data center solutions, according to supply chain analysis.
What to Watch
Capacity delivery timelines will determine whether these investments relieve bottlenecks or merely lock in allocations. Lumentum’s new U.S. fab is expected to come online in phases through 2027, while Coherent’s Pennsylvania expansion targets 800G and 1.6T module production by late 2026. Investors should monitor quarterly earnings for shipment volume disclosures and any commentary on yield rates for indium phosphide lasers, the component most prone to supply constraints.
Watch for similar moves by AMD, Broadcom, and custom-silicon hyperscalers. Microsoft, Google, and Meta are all designing optical interconnects for internal AI clusters and may pursue equity stakes or long-term supply agreements to secure capacity. If CPO adoption accelerates faster than pluggable transceivers, traditional optical vendors without advanced packaging capabilities risk margin compression.
Regulatory scrutiny could also emerge. Nvidia’s dual investments create preferential access that smaller AI Infrastructure players cannot replicate, potentially raising antitrust questions if optical component shortages constrain competitors. The Federal Trade Commission has previously examined exclusive supply arrangements in semiconductor markets; similar logic could apply to photonics if scarcity intensifies.