Opinion: optical transceivers at the chokepoint of AI growth and supply chain constraints

As AI infrastructure accelerates at an unprecedented pace, optical connectivity has become one of the defining enablers and constraints of next-generation data centers. In this Opinion Series, Yole Group explores how the optical transceiver ecosystem is evolving under the combined pressures of hyperscale AI demand, supply chain limitations, and shifting technology roadmaps.

Written by Yole Group’s photonics Senior Analyst, Martin Vallo, this article is part of our broader photonics intelligence offering designed to support strategic decision-making across the optical communications industry. To dive deeper into market forecasts, competitive dynamics, technology trends, and supply chain developments, discover Yole Group’s dedicated photonics collection, including Optical Transceivers for Datacom and Telecom 2026, Photonics Market Monitor – Optical Transceivers for Datacom, Co-Packaged Optics for Data Centers 2026, and more.

The optical transceiver industry has entered a new phase. For years, the market was driven by telecom cycles, hyperscale cloud upgrades, and predictable transitions from 100G to 400G. Today, AI clusters have changed the rhythm. Optical modules are no longer peripheral connectivity components; they are becoming one of the gating factors in how quickly AI infrastructure can scale. Yole Group reports that datacom optical component revenue exceeded $18 billion in 2025, up more than 70% year over year, with over 55 million high-speed 400G+ datacom modules shipped. 800GbE remains the growth engine, while 1.6T is already ramping and over 10 million 1.6T modules is forecast for 2026.

The problem is that the supply chain was not designed for this speed of acceleration. The final transceiver module is only the visible part of a much deeper chain: lasers, modulators, DSPs, TIAs, laser drivers, photodiodes, optical engines, connectors, fiber arrays, WDM filters, couplers, isolators, circulators, attenuators, and precision assembly and test. In the AI era, the bottleneck is less about assembling a box and more about qualifying thousands of tiny, yield-sensitive optical and electronic components at massive volume.

The most visible constraint today is the laser. Yole Group expects shipments of approximately 150 million 800G-and-above optical transceivers in 2026 and 2027, while warning that laser light sources have become a significant upstream bottleneck. The shortage is especially acute around EMLs, where capacity has reportedly been pre-allocated by strategic customers, extending lead times and forcing module makers and cloud service providers to seek alternative designs.

This is why the debate between EML and silicon photonics is no longer only technical. EMLs remain attractive because they provide proven reach, signal integrity, and manufacturability in today’s 800G and 1.6T pluggables. But the EML route depends on a narrow group of qualified InP-based laser suppliers. Silicon photonics shifts part of the value toward wafer-scale modulators and external CW-DFB lasers. In theory, this broadens the manufacturing base and improves integration. In practice, it simply moves the pressure point: from EML chip capacity to CW laser availability, silicon photonics foundry yield, known-good-die testing, optical coupling, and packaging.

The industry should therefore be careful with the phrase “InP shortage.” The issue is not simply the availability of raw InP wafers. The real shortage is qualified, high-yield, high-reliability compound-semiconductor capacity for 200G-per-lane optical links. Epitaxy, laser processing, hermetic or non-hermetic packaging, burn-in, and long-term reliability screening all matter. A marginally available laser is not the same as a hyperscale-qualified laser.

DSPs are the second pressure point. In conventional DSP-based pluggables, the DSP absorbs much of the signal-conditioning burden. That enables interoperability, reach margin, and operational robustness, but it adds cost, power, and dependence on advanced-node semiconductor supply. LPO challenges this model by removing the DSP from the module and using linear TIAs and drivers, while relying more heavily on the host ASIC and carefully controlled electrical channels. Marvell’s 1.6T LPO chipset, for example, combines 200G-per-lane TIAs and laser drivers for 800G and 1.6T linear-drive pluggable optics.

LPO will not eliminate DSPs, but it will segment the DSP market. For short, controlled, inside-the-rack or scale-up links, LPO can reduce module power and cost. For longer reach, heterogeneous environments, coherent links, 800ZR, 1.6T ZR/ZR+, and campus interconnect, DSPs remain essential. The consequence is not a collapse of the DSP market, but a reallocation: fewer PAM4 DSPs in some short-reach links, more value in coherent, coherent-lite, retimer, diagnostics, and host-side signal integrity.

Passive components are the underrated constraint. As bandwidth increases, the “passive” subassembly becomes less passive from a supply-chain perspective. Connectors, couplers, WDM mux/demux filters, isolators, circulators, attenuators, lenses, fiber arrays, and high-density optical interfaces must all meet tighter tolerances. In CPO, passive alignment, detachable fiber interfaces, pluggable laser sources, and fiber routing become system-level challenges. Broadcom’s CPO announcements highlight not only switch ASICs and optical engines, but also ecosystem partners for advanced fiber, connectors, sockets, pluggable laser-source cages, and high-density CPO fiber cables.

Geography is another bottleneck. China leads much of the high-volume module assembly and test ecosystem, with companies such as Innolight and Eoptolink central to 800G and 1.6T ramps. The U.S. leads in hyperscale architecture, switch ASICs, and DSPs, with several critical optical component suppliers. Neither side is fully independent. U.S. cloud and AI companies need Chinese manufacturing scale. Chinese module makers need access to advanced DSPs, lasers, and equipment, as well as global customers. This interdependence is becoming strategically uncomfortable.

The winners will be those with access to both technology and allocation. In modules, Innolight, Eoptolink, and Coherent are among the most important datacom players. Yole Group reported Innolight leading 2025 datacom revenue, followed by Eoptolink and Coherent. In lasers and optical components, Coherent, Lumentum, Broadcom, Mitsubishi, and Sumitomo remain key. In DSPs and high-speed electronics, Marvell, Broadcom, Cisco/Acacia, and selected ASIC players dominate. In CPO, Broadcom and NVIDIA are shaping the roadmap, with NVIDIA promoting silicon photonics switches that use fewer lasers and claim major power-efficiency advantages.

Looking forward, pluggables will not disappear quickly. They remain serviceable, interoperable, multi-sourced, and operationally familiar. Yole Group still expects no material impact on pluggable shipments from CPO over the next three years in its 2025 outlook. But CPO will grow where power density and bandwidth density justify the ecosystem risk: AI scale-up, high-radix switches, and, eventually, very large scale-out fabrics.

The supply chain will evolve from a module-centric model to a platform-centric model. The leaders will not simply be the companies with the cheapest transceivers: they will be the companies that can secure laser capacity, design, or source advanced DSP/driver/TIA silicon, qualify silicon photonics and EML alternatives, control passive alignment, and test, while satisfying both U.S. and Chinese geopolitical requirements. In optical transceivers, technology leadership and supply-chain leadership are now the same thing.

The optical transceiver market is entering a defining decade in which technological innovation alone will no longer guarantee leadership. In the AI era, success will depend equally on supply chain resilience, partnerships, manufacturing scalability, and geopolitical adaptability.

👉 Want to understand who will lead the next wave of optical connectivity? Explore Yole Group’s photonics reports and market monitor to access in-depth analyses, forecasts, and strategic insights shaping the future of AI-driven optical networks.

Martin Vallo, PhD, is Senior Technology & Market Analyst, Photonics at Yole Group.

Martin specializes in optical communication and semiconductor lasers within the Photonics & Sensing team at Yole Group. With 12 years’ experience in semiconductor technology, Martin is currently involved in the development of technology & market products as well as the production of custom consulting projects.

Prior to his mission at Yole Group, Martin worked at CEA (Grenoble, France), where he focused on the epitaxial growth of InGaN/GaN core-shell nanowire LEDs using MOCVD, and their characterization for highly flexible photonic devices.

Martin graduated from the Academy of Sciences, Institute of Electrical Engineering (Slovakia) with an engineering degree in III-nitride semiconductors.

As well as dedicated teardown analyses including optical transceivers and modules from NVIDIA, Huawei, and CISCO.

Source: www.yolegroup.com