Optical Transceivers for Datacom and Telecom 2023

Key Feature

• To provide the context of why DC operators and telecom carriers pay huge attention to optical interconnect evolution
• To understand the changes in DC network architecture
• Provide detailed market forecasts from 2017 to 2028 for optical transceivers in datacom and telecom
• Provide detailed market share analyses of the last three years split by datacom and telecom
• Review the supply chain for datacom and telecom
• Examine the application landscape, and associated technologies for datacom and telecom
• Review the optical transceiver industry with a focus on China

What’s new?

Review the trends toward higher speed and improved power efficiency

• Review the new modulator technologies TFLN, BTO, Organics and Graphene
• Review the potential and challenges of linear drive (no retimed) pluggable optical transceivers

Product objectives

• To provide the context of why DC operators and telecom carriers pay huge attention to optical interconnect evolution:
  • IP traffic growth, new applications such as market drivers, other trends in data centers impacting optical module technology, the market and industry, and the classification of fiber-optic communication and its technologies.
• To understand the changes in DC network architecture:
  • Network flattening trend: impact of switch radix
  • Fiber-to-the-server trend
  • Disaggregation DC – rack architecture clustering
• Provide detailed market forecasts from 2018 - 2028 for optical transceivers in datacom and telecom.
• Provide detailed market share analyses of the last three years, split by datacom and telecom.
• Review the supply chain for datacom and telecom.
• Examine the application landscape and associated technologies for datacom and telecom.
• Review the trends toward higher speed and improved power efficiency:
  • Review the new modulator technologies TFLN, BTO, organics, and graphene
  • Review the potential and challenges of linear drive (no retimed) pluggable optical transceivers
• Review the optical transceiver industry, with a focus on China.

Glossary

Glossary

Definitions: FPP vs. CPO

Definitions: Serdes

Definitions: Radix 

Report’s objectives

Scope of the report

Methodologies & definition 

About the authors

Companies cited*

What we got right, what we got wrong

Who should be interested in this report?

3-Page summary

Executive summary 

Context

• Global Trends in Datacom
• Global Trends in Telecom

Market forecasts 

Market shares & supply chain 

Datacom 

• DATACOM: Toward higher speeds  
• DATACOM: 800G and beyond
• DATACOM: Solutions for energy efficiency New materials for modulators
• DATACOM: Solutions for energy efficiency Linear Drive Optics
• DATACOM: Solutions for energy efficiency Co-Packaged Optics 
• DATACOM: Switch ASIC and SERDES

Telecom 

• TELECOM: Coherent DWDM
• Coherent Trend for Intra-DC & Inter-DC links: IM/DD vs Coherent 
• TELECOM: Passive optical networks 
• TELECOM: 5G optical networks 

Technology trends

Industry

• Chinese market and industry
• USA vs China

M&A / Partnerships  

Appendix

• Key  Parameters in Fiber-Optic Communication  
• From Telecommunication to fiber-optic communication 
• Optical Transceiver: 400G QSFP-DD DR4 
• Optical Transceiver: 100G QSFP28 CWDM

Yole Goup related products  

Accelink, Adtran, ADVA, Albis, Alibaba, Amazon Web Services, Applied Optoelectronics Inc (AOI), Arista, AT&T, Baidu, Broadcom (Foxconn Interconnect Technology, Avago), Broadex, China Mobile, China Telecom, China Unicom, Ciena (Cyan), CIGtech, Cisco (Acacia Communication, Luxtera), CloudLight, Coherent (II-VI, Finisar), ColorChip, Crealights, Credo, Dell EMC, E.C.I. Networks, EFFECT Photonics, Eoptolink, FiberHome, Fujikura, Fujitsu Networks, Fujitsu Optical Components, Furukawa, Gigalight, Global Communication Semiconductors, Google Cloud Platform, Guangdong Jiulian Technology, Guangxun Technology, Hengtong Group, HG Genuine Optics, Hisense Broadband, HiSilicon (Huawei), Huawei, Hyperlight, Infinera (Coriant, Transmode), InnoLight, Intel, Juniper Networks, Linktel, Liobate, Lumentum (Oclaro, NeoPhotonics, IPG Photonics (Menara Network)), Lumiphase, Luxshare, MACOM, Marvell (Inphi), Maxim Integrated, Meta, Microsoft Azure, Mitsubishi Electric, MOLEX (Oplink), NEC, Nokia (Alcatel Lucent, Elenion), NTT Electronics, NVIDIA (Mellanox), O-net, OE Solutions, Optical Library Technology, Ori-Chip Optoelectronics, Ranovus, Renesas (Integrated Device Technology), Rockley Photonics, Sicoya, Si-Photonics, Skorpios Technologies, Semtech, SoftBank, Source Photonics, Special message, Sumitomo Electric Device Innovations, Tacklink, TE Connectivity, Tencent, Verizon, YOFC, Zhongtian Technology, ZTE and more.

Generative AI will drive expanding data center infrastructure

The use of AI, led by technologies like Open AI’s ChatGPT and Google’s Bard, is gaining momentum and is driving demand in data centers. As technology integrators implement generative AI more deeply into everyday applications, the demand for computing power in data centers escalates. The server computer density that AI requires also generates significant heat, presenting energy efficiency and sustainability challenges. While AI is a primary factor driving demand for data centers, other applications, such as ultra-high-definition (UHD) video and a variety of augmented reality/virtual reality (AR/VR) applications and cloud services - social networking, business meetings, video streaming in UHD, e-commerce, and gaming, also continue to drive growth.

Revenue generated by the optical transceiver market increased 12.5%, from $9.8B in 2021 to $11B in 2022, and is expected to reach $22.2B in 2028 at a 12% compound annual growth rate for 2022-2028 (CAGR_2022-2028). This growth is driven by high demand for high-data-rate modules above 400G by big cloud service operators and national telecom operators requiring increased fiber-optic network capacity.

We anticipate the revenue growth rate for 2023 to be slightly negative. The main reason is that hyperscale CapEx is growing at an even lower rate. While the overall data center CapEx for the hyperscalers will decline significantly this year, telecom CapEx is expected to decrease up to 3% annually over the next three years. However, wireline and wireless telecom equipment could manage a small 1% growth rate year-on-year in 2023. One of the reasons for the slow growth of DC hyperscale CapEx is the shutting of the Metaverse and its negative consequences. Actions taken by Meta and the rest of the affected industry will result in a slowdown of optical transceiver deployment in 2023.

China’s position is becoming prominent driven by AI landscape

The global optical transceiver industry is highly competitive. The competitors range from large international companies offering a wide range of products to smaller companies specializing in narrow markets. The principal competitive factors in the optical transceiver market are the ability to provide leading-edge technologies for high-speed communications and to design and manufacture high-quality, reliable products, including customized solutions. Different techniques and approaches can be used to overcome physical limits to achieve higher data rates. It is clearly seen that the two strategies –  one based on InP and the other on Silicon Photonics platforms – will coexist during the coming years. Strong names are linked with both platforms.

Under its continuous catch-up and accelerated development program, China is taking an increasingly prominent position in the optical communication industry. Nowadays, the core optical technology for high-speed modules is the domain of American and Japanese manufacturers, but China has invested heavily in photonic manufacturing platforms – GaAs, InP, and SiPh. The Chinese government’s program – the Roadmap for the Development of the Optical Device Industry (2018-2022) – defined a national strategy to increase the market share of local producers of optical chips. The China-US trade restrictions and ZTE's ban may prompt China to increase its support for high-speed optical chips, and domestic optical chip production is expected to accelerate further.

AI-driven NVIDIA’s latest system design will require 800G pluggable optical modules for their upcoming AI/ML hardware. They are investing in linear non-retimed technology to alleviate limitations like interconnect power, power density, and latency for AI scaling. Top Chinese optics suppliers are preparing to be at the forefront of the suppliers. Their value in the stock market has tripled or quadrupled within a few last months.

It’s time to market for new designs with new materials

Silicon Photonics serves as a versatile platform that can host a wide variety of photonic components, such as modulators, photodetectors (PDs), splitters, (de)multiplexers, and filters, while it is limited in laser sources compared to more established III–V materials such as InP and GaAs. Recent years have seen a paradigm shift to where the integration of various materials has opened up additional routes toward performant SiPh modulators. The introduction of other materials such as TFLN (Thin Film LiNbO3), BTO (BaTiO3), plasmonic organic hybrid (POH), or graphene to the silicon platform for efficient phase shift promises alternatives to Si, relaxing trade-offs that constrain their power, performance (speed, drive voltage), and area.

Power consumption of high-speed optical interconnects is the main limiting factor for intra-data center optical links. The most significant contributors are the electrical interface SerDes between the switch ASIC and optical module, digital signal processors (DSPs), and additional retimers (CDRs), particularly for pluggable form factors. Co-Packaged Optics (CPO) is a new approach that brings the optics and the switch ASIC closer together and aims to overcome the abovementioned challenges. However, CPO must grow up to meet all the industry requirements. There is even a potential alternative to co-packaged optics in the form of Linear Drive Pluggable Optics (LPOs) remaining at the front panel of the switch. Reducing latency is a crucial improvement in applications such as switch-to-switch, switch-to-server, and GPU-to-GPU connectivity in machine learning (ML) and High-Performance Computing (HPC).