Optical transceivers and pluggable modules
BOTTLENECKInnoLight and three peers depend on EML lasers with 20+ week lead times, binding module supply for AI clusters.
Hot-pluggable modules converting electrical to optical signals at 400G, 800G, and 1.6T. Module supply constrains how quickly clusters can scale bandwidth. Top-4 vendors control ~70% AI market; EML laser availability is the binding bottleneck with 20+ week lead times.
Hot-pluggable electro-optical modules (QSFP-DD, OSFP, CDFP) that convert electrical signals to optical for inter-rack and long-reach connections at 400G, 800G, and emerging 1.6T.
Why the concentration exists
Pluggable optical transceivers are standardized modules that convert electrical signals to optical signals for high-speed data transmission. They serve as the interface between network equipment and fiber optic cabling, supporting speeds from 100 Gb/s to emerging 1.6 Tb/s switches. The modules are hot-pluggable, meaning operators can replace or upgrade them during service life without powering down the system. This standardized approach has made transceivers a cost-effective solution for scaling data center networks.[1][7]
Power consumption rises sharply with data rate, creating a physical constraint on module design and supply. An 800G pluggable transceiver consumes about 15 watts, while a 1.6T module using conventional technology could require 20 to 25 watts or more. Co-packaged optics alternatives, which integrate the optical engine directly with the switch ASIC, consume about 5.4 watts per 800G of bandwidth, a 65 percent reduction compared to pluggables. These power demands make thermal management a critical factor in module qualification and deployment.[2][8][21]
Optics have grown from a minor line item to the dominant cost in data center network hardware. At 10G speeds, optics represented about 10 percent of total network hardware cost, but at 400G and beyond, optics account for more than half of hardware spending. This cost shift has driven vertical integration and consolidation among suppliers, as controlling the optical component supply chain becomes essential to profitability in high-speed networking.[13]
What the evidence shows
InnoLight Technology holds the largest global market share for 400G and 800G modules to US cloud hyperscalers.
wolontek.comIn the 800G segment, Innolight's leading position faces narrowing gaps with Eoptolink, Coherent, and Nvidia.
m.c-light.comWho supplies it
Zhongji Innolight is the largest producer of optical transceivers in the world, making it a critical supplier for AI data center infrastructure. The company was formed in 2017 when Shandong Zhongji Electrical Equipment acquired InnoLight Technology and restructured the combined entity. InnoLight has demonstrated 1.6T OSFP-XD DR8+ optical transceivers and collaborated with Marvell on 800G ZR/ZR+ modules for data center interconnects. The company's silicon photonics transceivers are ramping from production facilities in Thailand.[14][15][17]
Multiple suppliers compete across different segments of the transceiver market. Broadcom supplies 100G to 800G optical transceiver modules for Ethernet switching, storage networks, and AI workloads. Cisco offers OSFP 1.6T and 2x800G modules for AI data center connections, along with 800G ZR/ZR+ coherent pluggable optics capable of transmitting up to 120 kilometers. LINK-PP provides 400G QSFP-DD, 200G QSFP-DD, and 800G transceiver modules, while Unitekfiber offers a portfolio spanning 10G SFP+ through 400G QSFP-DD modules.[12][6][3][18][23]
Component suppliers control critical inputs that limit module production capacity. Leading vendors including Coherent, MACOM, Broadcom, and Lumentum are releasing 200G photodiodes to enable 200G-per-channel data transmission. International players such as Coherent, Lumentum, and Applied Optoelectronics, along with Taiwanese firms Elite Advanced Laser Corporation and LuxNet Corporation, have initiated capacity expansions for electro-absorption modulated lasers and continuous-wave lasers. These component suppliers represent a concentrated upstream layer in the transceiver supply chain.[10][11]
Who controls it
What it depends on, and what depends on it
Electro-absorption modulated lasers and continuous-wave lasers are the primary components constraining transceiver production. Nvidia has secured capacity at key EML suppliers, which has contributed to extended lead times beyond 2027 and a worldwide shortage of these critical components. InnoLight's 400G optical transceivers based on Tower Semiconductor's PH18M silicon photonics process are in mass production, with 800G transceivers using the same process scheduled for volume production. The silicon photonics process enables higher integration but requires specialized fabrication capacity.[10][11][16]
Transceivers connect GPU clusters to network switches, carrying the east-west traffic generated during AI training workloads. A single AI training rack equipped with 16 H100 GPUs generates more than 400 Gbps of east-west traffic. Up to 33 percent of GPU time can be wasted waiting for network availability, representing more than $10,000 per GPU per year in idle compute costs. This dependency makes transceiver supply a binding constraint on AI cluster performance and utilization.[19]
Different transceiver form factors support varying reach distances for data center connectivity. The 800G SR8 and DR8 modules support short-reach connections up to 100 meters over multimode fiber. The 400G QSFP-DD DR4 modules support distances up to 500 meters over single-mode fiber, while 400G QSFP-DD LR4 modules extend reach to 10 kilometers or more. Coherent pluggable optics like the 800G ZR/ZR+ can transmit 800G wavelengths up to 120 kilometers with coherent ZR and over 1000 kilometers with OpenZR+.[20][12]
Where it sits in the stack
Takes in: EML/VCSEL laser chips (L4.6.5), PIC (L4.6.6), DSP/retimer ASIC (L4.6.7), module housing and connector
Sends on: 400G/800G/1.6T optical links for inter-rack and inter-pod fabric
What would break it
EML laser shortages have become the binding constraint on transceiver supply, with lead times extending beyond 2027. Nvidia's capacity reservations at key EML suppliers have tightened availability for other buyers. Component shortages of EMLs and continuous-wave lasers are identified as primary bottlenecks for capacity expansion. These lasers require specialized semiconductor fabrication processes and qualified supply chains that cannot scale quickly in response to demand spikes.[10][11]
Demand signal distortion compounds the supply constraint. When demand spiked for AI-driven applications, lead times that normally measured in weeks extended to months. Customers began double-ordering to secure supply, which further distorted demand signals and created artificial scarcity. This behavior amplifies the underlying component shortage and makes it difficult for suppliers to accurately plan capacity investments.[5]
Co-packaged optics technology threatens to displace pluggable transceivers in high-density applications. Nvidia has announced silicon photonics-based co-packaged optics switches that eliminate pluggable transceiver modules entirely. Google has demonstrated 40 percent power savings through optical circuit switching deployments. These architectural alternatives could reduce demand for pluggable modules in the largest AI data center deployments, though they require different supply chains and qualification processes.[4][2]
What to watch
The 1.6T optical transceiver generation is set to enter large-scale shipments in 2026, marking a transition point for the industry. Worldwide shipments of optical transceivers at 800G and higher speeds will reach 24 million units in 2025, then jump 2.6 times to nearly 63 million units in 2026. Volume deployments of 1.6T switches are projected for late 2026. These milestones will test whether the component supply chain can scale to meet demand at a new performance tier.[9][10][22]
Geopolitical factors could affect InnoLight's market position. In October 2025, the United States Department of Defense stated that Zhongji Innolight merits inclusion on a list of companies linked to China's military. This designation could restrict InnoLight's ability to supply US cloud hyperscalers, which have been the primary customers for its 400G and 800G modules. The outcome of any subsequent restrictions remains uncertain.[14]
Related nodes
Sources
- connectorsupplier.com · 2024-12-26T20:42:31
- newsletter.semianalysis.com · 2026-04-10T03:37:45
- unitekfiber.com
- introl.com · 2026-04-08T00:00:00
- 100gmodules.com · 2026-04-09T05:47:59
- resources.l-p.com · 2025-11-26T09:13:00
- omnitron-systems.com · 2026-03-04T00:00:00
- hytoptodevice.com · 2026-01-07T00:00:00
- digitimes.com · Apr 21, 2026
- eetasia.com · 2025-12-16
- trendforce.com · 20 April 2026
- cisco.com
- blogs.cisco.com · March 21, 2022
- en.wikipedia.org · 2017
- innolight.com · 2023-03-03
- eenewseurope.com · 4Q23
- prweb.com · 2024-03-20
- marketsandmarketsblog.com
- vitextech.com
- vitextech.com
- fibermall.com
- vitextech.com
- link-pp.com
Full scorecard, owner shares, supply edges and the full tracked roster are in the desk letter.
GET THE BRIEFING