/ From： / Review：5
In April 2026, the optical communications industry issued intensive signals of supply chain shortages, with contradictions between supply and demand erupting simultaneously across multiple sub-sectors.
Broadcom’s Director of Product Marketing recently warned in a media interview that the optical communications supply chain is facing production capacity bottlenecks. He acknowledged that over the past few years, the industry had widely regarded TSMC’s manufacturing capacity as “unlimited,” but today TSMC’s production constraints have become a major factor restricting the optical communications supply chain in 2026. Meanwhile, Japan’s Granopt, which holds a major global market share, announced a significant reduction in the production capacity of Faraday rotators, extending lead times for this critical component from the normal few weeks to 6–9 months. In the domestic market, price hikes in basic communications materials have also drawn notable attention. According to research data from China Post Securities, in early April 2026, the price of standard G.652D single-mode fiber soared from 18 RMB per fiber-kilometer to 85–120 RMB per fiber-kilometer, representing an increase of 450% to 567%. The spot market has even adopted an extreme pricing practice of “same-day validity,” where quoted prices expire the following day.
These recent market developments indicate that the optical communications industry is facing a structural supply shortage. While the upgrade of AI computing power networks underpins the surge in demand, current production pressures are concentrated in specific manufacturing segments and core material supplies, rather than an industry-wide absolute supply disruption.
TSMC Capacity Allocation and Packaging Bottlenecks
Broadcom’s supply chain warning directly targets the core logic chips in the optical communications hardware architecture. In modern high-speed optical modules, digital signal processors (DSPs) responsible for signal processing and switch chips in future co-packaged optics (CPO) solutions directly determine the quality and rate of signal transmission, and these core chips heavily rely on advanced semiconductor manufacturing processes.
In wafer foundry services, TSMC’s capacity allocation for advanced processes such as 3nm is under real pressure. Although optical communications-related DSP and ASIC switch chips demand cutting-edge process support, they are often at a disadvantage in the competition for capacity, with their allocated share squeezed by major customers including smartphone application processors (APs) and GPUs. Currently, the high-end DSP market is a duopoly between Broadcom and Marvell, whose advanced-process DSPs are highly dependent on TSMC’s production capacity.
In advanced packaging, TSMC’s CoWoS packaging capacity represents an even more critical bottleneck restricting the industry. Industry chain sources have confirmed that NVIDIA, backed by strong GPU demand, has reserved more than 70% of TSMC’s CoWoS-L advanced packaging capacity for 2025. This extreme capacity allocation structure creates an extremely high barrier for optical communications technologies such as CPO that require advanced packaging support to secure production capacity. Although TSMC plans to raise its monthly CoWoS capacity from 35,000 wafers in 2024 to 130,000 wafers by the end of 2026, such linear growth will still struggle to fully fill the market gap.
02
Beyond main semiconductor chips, severe production bottlenecks have also emerged in optical chips responsible for electro-optic conversion inside optical modules, as well as precision optical components and basic materials.
In 800G and 1.6T ultra-high-speed optical modules, electro-absorption modulated lasers (EML) and continuous-wave (CW) lasers serve as core devices enabling high-bandwidth, long-distance transmission. Amid the explosive demand for ultra-high-speed optical modules, global production capacity for EML and CW lasers has been nearly fully utilized. Leading manufacturers in North America, Japan and China are operating at full capacity yet still facing delivery pressures. Industry leaders such as Lumentum have explicitly stated that their backlog of orders for EML and CW laser chips has exceeded two years. For silicon photonics solutions, the supply of CW lasers further dictates the overall cost trend. Due to supply shortages, unit prices of certain CW laser specifications have even exceeded those of equivalent EML chips. Such supply-demand imbalances have created significant delivery shortfalls. To secure limited production capacity, North American cloud vendors typically hold stronger bargaining and premium-paying power, resulting in global scarce capacity being prioritized for the North American market.
Another factor exacerbating capacity constraints for optical chips is the shortage of upstream substrate materials. The manufacturing of high-end EML chips relies on indium phosphide (InP) substrates. Unlike silicon-based semiconductors, crystal growth of compound semiconductors such as InP is extremely difficult, with major challenges in yield control. Driven by the expansion of AI data centers, demand for indium phosphide is surging at an annual growth rate of 40%–50%. However, given the 18-month cycle for equipment qualification, InP substrate capacity cannot be rapidly expanded in the short term.
In terms of precision components, the Faraday rotator mentioned earlier is the core part of optical isolators, whose primary function is to prevent optical reflection from interfering with laser devices. The key reason behind Granopt’s production cut in Japan is the difficulty in securing raw materials: Faraday magneto-optical materials rely heavily on rare earths such as terbium gallium garnet (TGG). Following China’s implementation of rare earth export controls in January 2026, overseas manufacturers have faced significantly greater difficulties in obtaining critical rare earth materials.
Amid localized strains in the optical communications supply chain, computing power giants are investing heavily to build supply chain defenses. In March 2026, NVIDIA made three intensive strategic investments in core optical communications suppliers: injecting US$2 billion each into Lumentum and Coherent, with agreements including multibillion-dollar purchase commitments and locking in future production capacity rights for indium phosphide optical chips. Later that month, it invested another US$2 billion in Marvell Technology to strengthen interconnection capabilities for AI infrastructure.
In terms of technological evolution, co-packaged optics (CPO), while regarded as a long-term solution to power consumption and density challenges, has encountered obstacles in practical deployment. Since failures in data center optical modules mostly occur in the laser component, CPO’s design of integrating optical chips and ASICs on the same substrate means that a single laser failure could render an entire switch inoperable, posing significant cost risks. In contrast, near-packaged optics (NPO) places the optical engine and ASIC on separate substrates, allowing lasers to be replaced individually upon failure. With better maintainability, NPO is emerging as a more pragmatic choice for the industry at present.
Amid narratives of supply chain tightness, rational differentiation has also emerged in the market. Guosheng Securities noted that the current round of fiber price increases reflects expectations of future capacity shortfalls, representing a "supply-demand gap-driven" price surge. Spurred by such expectations, some enterprises and distributors have engaged in hoarding and speculative order picking, temporarily exacerbating spot market supply shortages.
The aforementioned industry pressure stemming from external supply shortages has objectively become a direct driver pushing China’s optical communications industry chain to extend into upstream core segments. Production capacity of overseas leading manufacturers has been locked up by giants, lead times have lengthened across the board, and some key materials even face supply disruption risks. This supply-demand imbalance in the global market has directly affected Chinese enterprises, forcing the accelerated validation and adoption of domestic supply chain solutions.
In the particularly scarce EML optical chip segment, domestic R&D and mass production are accelerating. According to QYResearch, the global market for high-speed EML lasers reached approximately US$654 million in 2025, providing ample room for domestic players. Chinese firms such as Source Photonics have achieved mass production of 25G EML chips, with revenue surging 138.5% in 2025. In early 2026, the company invested an additional 1.251 billion yuan to build Phase II facilities and expand capacity.
For precision components such as Faraday rotators, domestic enterprises are quickly filling the gap. In response to production cuts by Japan’s Granopt, domestic firms including CASTECH have begun small-batch deliveries of Faraday rotators and supporting magneto-optical crystals. Although such business currently accounts for only around 1% of their revenue, they have initially established capabilities comparable to U.S. and Japanese products. Leveraging China’s complete rare earth industry chain, domestic manufacturers are steadily strengthening their supply capacity in this segment.
Localized tightness in the optical communications supply chain is unlikely to ease fundamentally until new capacity comes online at scale in 2027. This predicament is pushing up the industry’s cost center, making long-term order locking the new norm, and shifting the focus of industrial competition from end-module assembly to upstream core chips and basic materials.
