Skip to main content

AI Data Center Race Tightens Optical Component Supply as Hyperscalers Secure Laser Diode Capacity

Published: 6.9.2026



Key Takeaways

  • NVIDIA, Google, and Meta are securing production capacity from EML and CW-DFB laser diode suppliers as AI data centers move toward 1.6 Tbps and higher-speed optical interconnects.
  • Combined monthly EML and CW-DFB laser diode capacity is expected to double to about 50.7 million units in 2026.
  • The supply-chain bottleneck for AI infrastructure is expanding beyond GPUs, HBM, and advanced packaging into optical components used in high-speed data center networks.


A New Pressure Point in the AI Supply Chain

For years, the AI infrastructure supply chain story centered on GPUs, high-bandwidth memory, and advanced packaging. Now, optical components are joining that list.


According to TrendForce, the rapid expansion of AI data centers is accelerating the industry's shift toward transmission speeds above 1.6 Tbps. As hyperscalers build larger GPU clusters and demand lower-latency connectivity at scale, NVIDIA, Google, and Meta have moved to secure production capacity from two critical laser diode suppliers: electro-absorption modulated laser (EML) and continuous-wave distributed feedback laser diode (CW-DFB LD) manufacturers.


The response from suppliers has been aggressive. Combined monthly production capacity for EML and CW-DFB laser diodes is expected to roughly double to around 50.7 million units in 2026. The top three suppliers across both categories, Broadcom, Lumentum, and Sumitomo Electric, are projected to account for 55% of that capacity.


Why EML and CW-DFB Laser Diodes Matter

Faster AI infrastructure is about moving data quickly between those processors across GPUs, switches, racks, and facilities. That's where optical interconnects become critical.


EMLs are high-performance components that integrate a laser source and an electro-absorption modulator on a single chip. Because the laser diodes remain continuously active, EMLs offer ultra-low noise interference and narrow spectral characteristics, making them well-suited for data transmission exceeding 800 Gbps and for medium- to long-distance links beyond 2 kilometers. Given the technical complexity involved, the EML market is highly concentrated: Lumentum, Broadcom, and Mitsubishi Electric collectively hold approximately 72% market share.


Lumentum, in particular, is pushing the technology envelope. The company is expanding production capacity for 100/200 Gbps EML products and demonstrated 400 Gbps-per-lane EML technology at OFC 2026, targeting future 3.2 Tbps market demand.


CW-DFB laser diodes are gaining traction along a different path. While NVIDIA-led EML solutions focus on signal integrity and transmission performance, other cloud service providers are actively developing CW-DFB LDs for optical circuit switches and silicon photonics co-packaged optics solutions. Broadcom and Sumitomo Electric currently lead CW-DFB capacity, followed by Coherent and LandMark/LuxNet, collectively accounting for approximately 74% of total CW-DFB market capacity.


Both technologies are essential to the 1.6T transition. And neither can be scaled overnight.


NVIDIA's $4 Billion Signal

On March 2, 2026, NVIDIA announced separate multiyear strategic partnerships with Coherent Corp. and Lumentum Holdings, each including a $2 billion investment alongside multibillion-dollar purchase commitments and future capacity access rights.


The Lumentum agreement supports R&D, future capacity, and operations as Lumentum builds out U.S.-based manufacturing capabilities in a new fabrication facility. The Coherent agreement similarly supports research and development, future capacity, and operations as Coherent expands its domestic manufacturing footprint.


Both agreements are non-exclusive, leaving Lumentum and Coherent to continue serving other customers. But NVIDIA's capacity commitments are substantial enough to reshape availability. TrendForce reports that NVIDIA's strategic lock-in of EML supplier capacity has pushed delivery lead times beyond 2027, creating a supply shortage in the laser light source market.

NVIDIA has also embedded optical technology directly into its networking platforms. Its Spectrum-X Photonics and Quantum-X Photonics switches use silicon photonics and co-packaged optics to improve power efficiency, signal integrity, and system resiliency at scale.


A High-Growth Market With Tightening Supply

The laser diode capacity picture is part of a broader surge in AI optical infrastructure spending. TrendForce projects the global market for AI-focused optical transceivers will expand from US$16.5 billion in 2025 to US$26 billion in 2026, representing more than 57% year-over-year growth.


This growth reflects not only specification upgrades but a broader structural reshaping of the optical communications supply chain driven by accelerating AI data center deployment. Traffic at North American hyperscale data centers has sustained over 30% annual growth, prompting cloud providers to expand GPU and AI server deployments and increase procurement of high-speed optical interconnects.


Shipments of 800G and higher transceivers are forecast to surge from 24 million units in 2025 to nearly 63 million units in 2026, a 2.6-fold increase in a single year.


Supply, however, is not keeping pace cleanly. TrendForce identifies several compounding constraints: tight availability of EML and CW laser chips, the difficulty of scaling high-precision optical alignment processes, and rising power consumption and thermal management demands that add complexity to both module design and data center deployment.



Concentration Creates Allocation Risk

The structure of this market amplifies the supply pressure. With the top three EML suppliers holding roughly 72% of capacity, and the top CW-DFB players accounting for around 74%, a limited number of companies have outsized influence over how quickly AI optical networks can scale.


When hyperscalers secure production through long-term agreements, as NVIDIA has done, the open market tightens for everyone else. Companies without comparable purchasing scale may face longer lead times, constrained allocation, and reduced flexibility in spot-market procurement.


This risk extends well beyond AI infrastructure. Telecom networks, enterprise data centers, industrial automation, defense applications, and high-performance computing platforms all draw from the same optical component pool. Pressure at the top of the demand stack tends to ripple outward.


Stay up to date
Read industry news, product offers, and events.
Join email list