The optical communications industry historically exhibited clear telecom infrastructure cycles. However, with the rapid surge in intra-data center data transmission demand, the industry is shifting toward an AI-driven structural growth phase. As AI cluster architectures continue to expand, demand for high-speed optical interconnects is rising significantly. 800G is expected to enter large-scale deployment between 2026 and 2027, while 1.6T technology is already being introduced, further driving strong demand for high-speed Ethernet modules, silicon photonics, and co-packaged optics (CPO).
According to LightCounting’s latest report released in January 2026, the share of optical components in capital expenditures by the top five cloud service providers is projected to increase from 2.7% in 2025 to 3.1% in 2026, and further to 4.1% by 2031. This trend indicates that optical communication is becoming a core component of AI computing infrastructure, especially in supporting scale-out and scale-up networks within AI clusters. High-speed optical modules, CPO, and higher-bandwidth interconnect technologies are gaining importance.
From a supply chain perspective, U.S. companies continue to dominate high-value segments, including optical materials, InP and silicon photonics technologies, PAM4 optical DSP chips, and AI data center switching and Ethernet architecture design. Representative companies such as Marvell, Coherent, and Arista are actively positioning themselves in the high-speed optical interconnect upgrade cycle. In contrast, Taiwanese companies mainly focus on manufacturing segments such as optical transceiver assembly, subsystem integration, optical component production, and packaging/testing, playing a key role in mass production support for 800G and 1.6T module upgrades.
U.S. Optical Communications Industry Chain
The optical communications industry features a highly specialized division of labor.
- Upstream players provide core optical and electrical components, including optical materials, laser chips, photodetectors, high-speed signal processing chips (DSP/SerDes/Driver), and integrated components such as silicon photonics PICs and optical engines.
- Midstream players—optical transceiver manufacturers—integrate these components into deployable products such as transceivers, AOCs, or DACs, completing design, packaging, assembly, and testing.
- Downstream, switch and router vendors deploy these products into data center network architectures, while CSPs and AI data center operators purchase and deploy them at scale to support AI training, inference, and high-speed data transmission with high bandwidth, low latency, and low power consumption.
| Industry Segment | Representative Companies (U.S.-listed) | Industry Overview | Revenue Model |
|---|---|---|---|
| Testing and Measurement Equipment | TER, KEYS | These companies mainly provide testing and measurement equipment used in the R&D, validation, and mass production stages of optical communications and high-speed data transmission components, modules, and systems. This includes wafer testing, component testing, module testing, signal integrity analysis, bit error rate testing, as well as bandwidth and power consumption validation. As optical module speeds continue to advance, requirements for high speed, high frequency, low latency, and high reliability also rise, making testing equipment critical to ensuring yield, performance, and production stability. | Revenue is mainly generated through the sale of testing equipment, measurement instruments, and automated test systems. Customers include chipmakers, optical component manufacturers, module makers, and system equipment vendors. Some companies also provide maintenance and calibration, software upgrades, consumable replacement, and technical services. |
| Optical Materials and Basic Components | GLW, COHR | These companies provide the foundational optical components and materials required for optical communication modules, such as optical fiber, glass substrates, filters, lenses, optical coatings, and packaging materials, placing them at the very upstream end of the industry chain. Although these products are generally lower-priced than high-end chips and modules, they play a critical role in optical loss, stability, heat dissipation, and signal quality, and continue to upgrade alongside rising demand for high-speed transmission and high-density packaging. | Revenue is generated through shipments of materials or components, with income mainly determined by shipment volume, specifications, and yield rates. Some higher-end products can maintain better margins through customized specifications, long-term supply agreements, and technical barriers. |
| Emitter Chips and Lasers | LITE, COHR | These components convert electrical signals into optical signals. Core products include laser emitters such as EML, VCSEL, and DFB, which are widely used in short-, medium-, and long-distance data transmission scenarios. As transmission speeds upgrade from 400G to 800G and even 1.6T, requirements for laser output efficiency, stability, power consumption, and heat dissipation also increase, making them a key foundation for the performance of high-end optical modules. | Revenue mainly comes from sales of emitter chips or laser components. Customers are typically module manufacturers or integrators. Revenue is closely tied to data center upgrade cycles, penetration of high-speed products, and customized design projects. |
| Receiver Chips and Photodetectors | AVGO, COHR, LITE | These components convert optical signals back into electrical signals. Core products include photodetectors such as PIN and APD, along with related receiver-side chips. Receiver performance directly affects transmission sensitivity, bit error rates, and overall module stability, making it especially important in high-speed, long-distance, or high-density transmission environments. | Revenue is generated through shipments of receiver components and photodetectors, often supplied together with transmitter-side components to module makers. Higher-end products can enjoy better ASPs due to upgraded specifications and stricter reliability requirements. |
| Core Electrical Chips in Optical Modules (Signal Processing): DSP / SerDes / Driver / TIA | MRVL, AVGO, MTSI | These four categories are all core electrical chips within optical modules. As high-speed data center standards become more widely adopted, chip complexity has increased significantly, making this one of the industry chain’s higher-value and more technically demanding segments. Driver: converts conditioned electrical signals into current signals capable of driving lasers to emit light. DSP: compensates for and corrects high-speed signals. SerDes: consolidates multiple slower parallel electrical signals into a smaller number of high-speed serial signals for transmission, then converts them back into multiple data streams at the receiving end. TIA: amplifies extremely weak current signals into electrical signals that the DSP can process. | Revenue mainly comes from chip sales, with ASPs varying based on specifications, speed, channel count, and level of integration. Some suppliers can also increase revenue and customer stickiness through custom ASICs, joint development, and platform-based solutions. |
| Silicon Photonics PIC (Optical Path Processing) | INTC, MRVL, CSCO | Silicon photonics PICs (Photonic Integrated Circuits) are core components that integrate multiple optical functions—such as modulation, splitting, combining, optical path transmission, and signal coupling—onto a single chip. This improves integration, reduces size, and lowers power consumption in optical modules. Compared with traditional architectures made up of discrete optical components, silicon photonics is better suited to high-speed, high-density, and large-scale manufacturing needs, making it an important technology driving high-speed, low-power optical communications upgrades in AI data centers. | Revenue sources include silicon photonics chip sales, joint development projects with module manufacturers or cloud customers, and in some cases technology licensing or platform integration income. In the early stages, revenue is often driven by design wins, while larger-scale shipments expand revenue later on. |
| Optical Engines | COHR, CSCO, MRVL | Optical engines are subsystems that highly integrate lasers, PICs, driver components, and receiver components. They can be viewed as the core optical building block within an optical module, helping improve signal integrity, reduce size, and lower power consumption. As switch bandwidth continues to increase and new architectures such as CPO develop, the importance of optical engines is steadily rising. | Revenue mainly comes from shipments of integrated solutions to module manufacturers, equipment vendors, or certain large customers. Product unit prices are usually higher than those of individual parts, and revenue is closely tied to upgrade cycles in high-end networking equipment and the pace of adoption for new architectures. |
| Optical Transceivers | COHR, AAOI, LITE | Optical transceivers are the core midstream products in the optical communications industry chain. They integrate transmit and receive components, DSPs, PCBs, thermal management, and packaging design into pluggable transceiver modules such as 100G, 400G, 800G, and 1.6T. They directly benefit from data center bandwidth upgrades and AI cluster expansion, and have been one of the fastest-growing areas in optical communications in recent years. | Revenue is mainly generated through sales of module products. Customers include switch equipment vendors, cloud service providers, and system integrators. Revenue is typically influenced by transmission speed upgrades, ASP changes, shipment volume, and capital expenditure cycles at major customers. |
| Active Optical Cables (AOC) / Direct Attach Copper (DAC) | APH, TEL, AAOI | DAC: uses twin-ax copper cables to transmit electrical signals directly, without optical components, offering advantages in lower cost and lower power consumption, though transmission distance is limited. AOC: packages VCSEL lasers, photodetectors, and driver ICs at both ends of the cable, converting electrical signals into optical signals for transmission through fiber, providing long-distance capability, better interference resistance, and high bandwidth. In application, scale-up architectures can continue using DAC alongside the development of 200G/400G SerDes to reduce power consumption and deployment costs while improving cost performance. In contrast, scale-out and DCI architectures must rely on fiber-based solutions (AOC/optical modules) to meet long-distance, low-latency transmission requirements. | The revenue model is mainly based on one-time hardware sales driven by shipment volume. DAC is highly standardized and therefore more price-sensitive, with margins easily pressured by copper prices and competition. AOC, by contrast, is priced according to speed, distance, and certification, carries a technical premium, and is often sold together with switches and GPU servers. |
| Switches and Routers | ANET, CSCO | Switches mainly operate within local area networks (LANs), handling high-speed packet forwarding and traffic scheduling. They are essential for achieving high-bandwidth, low-latency interconnection inside data centers, such as in leaf-spine architectures. Routers, on the other hand, handle packet path selection and forwarding between different networks, and are used in enterprise wide area networks (WANs), telecom backbones, internet core networks, and global data transmission and connectivity. As cloud computing, AI, and 5G continue to develop, both are evolving into programmable network platforms that integrate security, traffic engineering, and automation. Demand from AI and cloud computing is also driving continued growth in the high-speed switch market, including 800G and 1.6T. | The revenue model is centered on one-time hardware sales (CapEx), while gradually shifting toward a hybrid model of “hardware + software + services.” Vendors generate recurring revenue through NOS licensing, subscriptions, and maintenance services. |
| AI Data Centers / CSPs | MSFT, GOOGL, AMZN, META | AI data centers are hyperscale computing infrastructures built around GPUs/TPUs and high-speed networks to support AI training and inference. CSPs provide computing, storage, networking, and AI capabilities through large-scale data centers, making them core infrastructure for generative AI and enterprise digital transformation. As AI workloads surge, around 60–65% of related demand is expected to be carried by these CSPs. | Capital-intensive data centers are converted into “on-demand + subscription-based” services. Revenue comes from IaaS, PaaS, SaaS, and the rapidly growing AI-as-a-Service segment, with customers paying based on usage or subscription. This allows one-time CapEx to be transformed into long-term, scalable OpEx revenue, while economies of scale help improve margins. At the same time, data gravity and high switching costs strengthen customer stickiness. |
Industry Outlook
- Continuous improvements in GPU computing power and switch throughput are increasing the demand for inter-node transmission speeds in AI data centers, driving optical transceivers toward higher density, bandwidth, and speed. As specifications evolve from 800G to 1.6T, both transceivers and components such as lasers will undergo upgrades.
- Silicon photonics (SiPh) is currently mainly used in high-end optical transceivers. Its future adoption path is expected to evolve from Transceivers → CPO switches → Optical I/O. With upgrades to 3.2T transceivers around 2027–2028 and the commercialization of 200G-per-lane silicon photonics lasers, SiPh penetration in AI data centers is expected to grow from less than 15% in 2023 to over 47% by 2028, becoming a mainstream solution.
- As transmission speeds increase, signal loss between optical transceivers and switches becomes more severe. CPO can significantly reduce power consumption and latency. 2026–2027 will be a key commercialization phase for CPO, initially adopted in scale-out switch architectures. It is expected to gain significant penetration at 200G per lane (1.6T) and become dominant at 400G per lane (3.2T), with large-scale deployment likely between 2028 and 2030.
- All-Optical Networks (AON) will accelerate as bandwidth demand, power constraints, and system complexity rise. Adoption will begin at the transmission and data center interconnect layers, gradually expanding to switching and computing layers. With silicon photonics and CPO, networks will shift from electro-optical hybrid architectures to primarily optical systems, addressing power and bandwidth bottlenecks.
2026 Catalysts
- The four major CSPs (Microsoft, Google, AWS, Meta) continue to raise capital expenditures, expected to exceed $700 billion in 2026. Investments in AI servers and data centers will drive strong demand for optical transceivers.
- Increasing AI inference demand is boosting shipments of ASICs and GPUs, along with upgrades in AI chip specifications. This drives higher demand for optical transceivers. In 2026, demand for 800G transceivers is expected to exceed 40 million units, while 1.6T demand is projected above 30 million units. Supply-demand imbalances may lead to capacity expansion and higher ASPs.
- Looking ahead to 2026–2027, rising CoWoS capacity indicates sustained AI investment. EML supply shortages are projected at 36% in 2026 and 48% in 2027, supporting price increases and revenue growth.
- CSPs are increasingly developing in-house AI models. Despite customization in ASICs and AI models, transmission speeds are constrained by current laser and DSP technologies. As 1.6T transceivers and switches enter mass production in 2026, they are expected to become a key competitive focus among CSPs.
Potential Risks
- Geopolitical risks may restrict Chinese optical companies’ access to key components and markets, potentially causing supply mismatches and cost increases, especially given their strong share in 400G/800G modules.
- Laser production faces bottlenecks due to long validation times (2,000–5,000 hours) and equipment constraints. Failure in customer validation can lead to prolonged shortages and delayed shipments, increasing industry concentration.
- Broadcom noted tight capacity in lasers and wafers, with expansion not expected until 2027. PCB lead times have extended to six months, and supply chain constraints may delay AI chip shipments and slow optical adoption.
- InP substrate supply is constrained due to high demand, concentrated production, and geopolitical risks. High technical barriers and long expansion cycles make it a key bottleneck for industry growth.
Conclusion
The optical communications industry spans upstream optical components and electrical chips—such as lasers, photodetectors, DSP, SerDes, Driver, and TIA—extends through midstream optical engines, silicon photonics PICs, and transceiver integration, and ultimately serves downstream applications including switches, routers, AI data centers, and cloud platforms.
Driven by AI training and inference demand, as well as continued CSP capital investment, the industry is expected to maintain strong growth in 2026, accelerating the adoption of next-generation technologies such as 800G, 1.6T, silicon photonics, and CPO.
However, despite a positive outlook, investors should closely monitor supply bottlenecks, delays in technology validation, fluctuations in customer demand, and geopolitical and capital expenditure risks.
