AI Optical Interconnect Market Size, Share & Growth Forecast (2026-2031)

The AI Optical Interconnect Market is projected to grow from approximately USD 26.0 billion in 2026 to USD 79.0 billion by 2031, at a CAGR of 24.89% during the forecast period.

The market includes optical transceiver modules designated for AI computing clusters, the optical portion of integrated and co-packaged networking products, optical circuit switching equipment for AI fabrics, and dedicated coherent optical links between AI data centers.

Optical interconnects reduce the need to carry increasingly high-speed data over long copper connections. A modern AI network uses short electrical connections where appropriate but relies on optics for distances, port densities and bandwidth that copper cannot achieve at acceptable power or signal-integrity cost. Silicon photonics, indium phosphide laser technology and advanced packaging provide alternative routes to higher-capacity products.

Training large models and serving many inference requests require tightly coordinated networks rather than isolated accelerator servers. When thousands of devices exchange gradients, intermediate activations or memory-resident data, networking capacity can determine how much useful compute the operator obtains from the system. Optical modules and photonics-enabled switches provide the inter-rack bandwidth necessary to support these high-intensity data flows.

AI-oriented optical links currently span several equipment classes. Pluggable modules with 800G interfaces serve widespread cluster requirements, while 1.6T products increase per-port bandwidth and allow operators to expand fabrics with fewer physical interfaces. Marvell expanded its 1.6T DSP portfolio in March 2026 and described its Ara technology as shipping in mass volume. Coherent has demonstrated multiple 1.6T transceiver approaches, including silicon photonics, InP and VCSEL devices, reflecting competing design choices.

Optical circuit switching introduces a different method of moving traffic by establishing direct optical paths without repeated electronic packet processing for selected connections. Lumentum and Marvell demonstrated interoperability between an OCS platform and several optical module classes in March 2026. NVIDIA and Broadcom meanwhile are moving optical engines inside switching packages, potentially reducing the electrical distance between the switch ASIC and optical conversion.

Market Trends

• 800G and 1.6T Modules Become Core AI-Fabric Building Blocks

Growing accelerator populations require more bandwidth between racks without multiplying cables and transceiver ports at the same rate as cluster size. The transition from 400G to 800G and 1.6T optical links allows cloud operators to connect high-density compute and switching infrastructure while addressing severe space, power and fiber management constraints.

Marvell announced in March 2026 that its 1.6T Ara optical DSP was shipping in mass volume to global customers. Coherent has demonstrated multiple 1.6T pluggable transceivers using alternative laser and photonic technologies. Module availability remains dependent on qualified laser chips, optical packaging yields and reliable operation across many ports, making supplier delivery commitments as important as specifications.

• Co-Packaged Optics Changes the Network Equipment Architecture

At extremely high switch throughput, repeated electrical paths between a switch ASIC and front-panel transceivers consume power and create signal losses. Co-packaged optics moves optical engines close to the switch silicon and can reduce the amount of long, power-hungry electrical signaling, although it also introduces serviceability and thermal-management challenges.

Broadcom announced shipments of its third-generation Tomahawk 6 Davisson CPO switch in October 2025, while NVIDIA stated in May 2026 that Spectrum-X Ethernet Photonics switches were in production. The ability to replace failed optical elements and manage integrated lasers will influence customer economics, so CPO is likely to advance alongside rather than instantly replace pluggable modules.

• Optical Circuit Switching Adds a Reconfigurable Fabric Option

Optical circuit switches can move selected large traffic flows through direct paths and reduce repeated optical-electrical-optical conversion. These systems are attractive for workload patterns with predictable or controllable communication, particularly where large-scale accelerator networks require periodic topology changes or high-bandwidth groupings.

Marvell and Lumentum demonstrated a shared optical circuit-switching arrangement at OFC 2026, combining the Lumentum switching platform with Marvell-powered intra-data-center and longer-distance modules. Adoption will depend on scheduling, network control software and whether circuit utilization benefits outweigh the additional operational complexity.

• Scale-Across AI Training Brings Coherent Optics Into Cluster Planning

Large AI operators increasingly distribute workloads and associated storage across multiple facilities. Connecting separate data centers requires optical transport with greater reach and stronger signal management than short data-center links, introducing coherent ZR/ZR+ products into AI-network planning and broadening the role of optical vendors beyond conventional rack-to-rack traffic.

Marvell expanded its COLORZ coherent product range in March 2026, introducing 1.6T ZR/ZR+ hardware and advanced optical DSP technology. These announcements reflect a clear direction toward higher-capacity scale-across networks, but recently introduced products must pass customer qualification before their announced capabilities can be treated as a mature installed base.

AI Optical Interconnect Market Segment Analysis

By Commercial Offering

• AI-Cluster Pluggable Transceivers

Pluggable optical transceivers account for approximately USD 21.0 billion in 2026 and are projected to reach USD 55.3 billion in 2031. Products at 800G currently carry much of the demand, while 1.6T modules support the rising bandwidth requirements of dense accelerator systems. Large deployments often specify both the link technology and module supplier qualification to maintain stable performance at the required distances.

Coherent, InnoLight, Eoptolink and Applied Optoelectronics participate in the module supply chain, while Marvell contributes optical DSP technology used in qualified modules. Revenue is most appropriately assessed at the finished optical interconnect equipment level rather than adding every DSP, laser and module invoice together. Availability of efficient lasers and high-yield packaging remains a major differentiator.

By Optical Integration Architecture

• Pluggable Modules

Pluggable transceivers represent an estimated 83% of 2026 value because operators can replace modules and upgrade bandwidth independently of the switch chassis. This is attractive when different racks and fabrics use separate reach requirements, such as short-reach intra-cluster links and longer inter-building connections.

By 2031, the estimated pluggable share is expected to decline toward 70% as co-packaged optical engines, near-package optics and optical switching gain commercial traction. Broadcom and NVIDIA show that integrated optics can be shipped with high-capacity switches, but operators must weigh lower link power against replacement economics and maintainability.

By Link Transmission Rate

• 800G

800G optical links contribute an estimated 60% of revenue in 2026, reflecting broad data-center adoption and practical deployment maturity. This transmission class supports dense clusters while 1.6T qualification expands across DSPs, lasers and transceiver packaging. Revenue share differs from port shipment share because higher-speed modules generally carry different selling prices.

The model assigns approximately 35% of 2031 sales to 800G and another 35% to 1.6T, with 3.2T and above increasingly important late in the forecast period. Coherent has publicly demonstrated next-generation configurations, but demonstration speed should not be interpreted as immediate broad customer procurement.

By Primary Network Role

• Scale-Out Cluster Links

Scale-out optical links account for approximately 66% of 2026 market value because they connect compute racks, storage systems and data-center network switches used by large AI clusters. Reliable high-bandwidth communication across these domains reduces network bottlenecks and protects the utilization of expensive accelerators.

The model also anticipates a larger scale-up optical role as multi-rack accelerator domains increase. New optical scale-up specifications and Marvell Photonic Fabric technology point to a longer-term market in which optical links connect processing and memory resources that would previously have remained within an electrical domain.

By Photonics Technology

• Silicon Photonics

Silicon photonics accounts for an estimated 39% of market value in 2026 and is expected to expand to 48% by 2031 as dense optical engines become more common. Integrating modulators and waveguides into photonic chips can improve packaging density and help vendors manufacture high-speed interfaces at scale.

InP and externally modulated laser technologies remain essential, particularly for high-performance links and efficient laser sources. Coherent has shown silicon photonics, InP and VCSEL approaches in parallel, confirming that technology selection remains specific to reach, optical power budget and customer qualification rather than converging on one universal design.

By Buyer and Deployment Type

• Hyperscale AI Cloud

Hyperscale cloud infrastructure operators represent an estimated 61% of 2026 market value. Their large training and inference clusters require predictable module supply, support for repeated platform generations, and a credible path from 800G to higher-capacity links. Procurement is concentrated among a small set of large buyers that can shape standards and qualification requirements.

Neocloud and GPU-cloud operators are modeled at 16% of demand in 2026, with a rising share as specialized providers expand capacity. Sovereign and enterprise AI centers form another route to demand, though their purchasing cycles and module qualification processes can differ significantly from the largest hyperscalers.

By End-User Industry

• Cloud Computing and Model Developers

Cloud computing and AI model developers represent about 66% of 2026 market value because foundation-model training and large-scale inference create substantial accelerator-to-accelerator traffic. These buyers often prioritize network stability, power efficiency and predictable transceiver procurement over individual component prices.

Telecom and network service providers contribute through coherent inter-data-center links, while financial services, industrial enterprises and research organizations expand demand for smaller AI clusters. Their procurement needs vary by workload and site distribution, so suppliers will need both hyperscale-qualified products and flexible enterprise configurations.

Market Drivers

• Expansion of Multi-Rack AI Training Clusters

AI training clusters exchange gradients and intermediate data across thousands of devices, increasing the bandwidth needed between racks. Optical transceivers and photonics-enabled fabrics help maintain accelerator utilization where conventional electrical links become impractical because of reach, power and signal constraints.

• Commercial Migration to 800G and 1.6T

Operators are upgrading high-speed fabrics as 800G deployment becomes established and 1.6T platforms mature. Marvell reports mass-volume shipments of 1.6T optical DSPs, while transceiver suppliers are introducing qualified optical engines that support greater bandwidth per switch port and improve rack connectivity density.

• Pressure to Reduce Optical Link Power Consumption

Large AI installations contain substantial numbers of optical interfaces, so improvements in watts per transmitted bit translate into meaningful operating-cost savings. The production CPO initiatives from NVIDIA and Broadcom show how integrated optical conversion can help limit signal losses around high-capacity switching silicon.

• Optical Circuit Switching and Topology Reconfiguration

Changing AI workload patterns create opportunities for switches that form optical paths between different groups of equipment. Lumentum and Marvell have demonstrated OCS interoperability, indicating the potential for operators to reduce avoidable electrical conversions and improve the utilization of selected high-capacity connections.

• Geographically Distributed AI Infrastructure

Cloud operators increasingly plan multiple AI locations to balance power availability, customer demand and infrastructure resilience. Coherent optical DCI supports long-reach movement of model data and storage traffic between facilities, extending high-speed optical demand beyond individual compute halls.

Market Restraints

• Advanced Laser and Packaging Supply Constraints

High-speed transceivers depend on specialized lasers, photonic integrated circuits and tightly controlled packaging. Limited yields or shortages in any of these inputs can postpone AI cluster acceptance and cause operators to maintain a mix of older and newer link generations for longer than planned.

• Power and Thermal Limits for Integrated Optics

Optical integration can lower electrical link losses, but placing engines close to a hot switch ASIC changes cooling and reliability requirements. Customers need assurance that laser sources, optical connectors and packaged devices will remain serviceable in densely populated data-center environments.

• Complex Supplier and Module Qualification

An AI fabric may require thousands of optical links to perform consistently across operating temperatures and fiber distances. Buyers therefore test interoperability, bit-error performance and failure handling before approving new suppliers, increasing the time between a product demonstration and meaningful recurring revenue.

• Continuing Competitiveness of Short-Reach Copper

Copper still provides attractive cost and operational simplicity across short distances inside racks or chassis. Optical interconnect products must therefore demonstrate a strong power, reach or bandwidth benefit rather than assume that every accelerator link will be converted to fiber.

• Rapid Technology Turnover and Price Pressure

Moving from 800G to 1.6T and eventually higher speeds introduces significant development expense and risks shortening the commercial life of installed modules. Suppliers also face price pressure from large customer contracts and competition among silicon photonics, InP and VCSEL architectures.

Regional Outlook

• North America

North America accounts for an estimated 48% of market value in 2026 and is projected to reach approximately USD 34.8 billion by 2031. The United States contains a high concentration of hyperscale cloud operators and AI model developers purchasing large-scale accelerator clusters. NVIDIA, Broadcom, Marvell and optical component suppliers contribute to the technology ecosystem, while operators specify demanding requirements for bandwidth, power and network availability.

The region is expected to lead early co-packaged optics adoption because large buyers can integrate new switch architectures into planned AI facilities and qualify integrated optical systems at scale. Mature procurement practices, long deployment schedules and supplier qualification requirements will shape adoption. High-speed module shipments and photonics-enabled switching will remain important sources of spending even as network designs evolve.

• Asia Pacific

Asia Pacific is expected to be the fastest-growing regional market, increasing from approximately USD 6.8 billion in 2026 to USD 22.9 billion in 2031. China, Taiwan, Japan and South Korea combine optical component manufacturing, semiconductor design and expanding demand for AI hardware. The region supplies key laser, transceiver assembly and packaging capabilities supporting global AI cluster investment.

Regional suppliers such as InnoLight and Eoptolink participate in high-speed optical module markets, while Korean and Taiwanese semiconductor ecosystems support the transition to advanced optical packaging. Growth will depend on local cloud and sovereign AI deployment alongside export demand. Supply bottlenecks and international procurement restrictions may alter the balance between production locations and ultimate equipment installation.

Competitive Landscape

The AI optical interconnect market brings together transceiver manufacturers, optical component specialists, networking-platform vendors and merchant semiconductor providers. NVIDIA and Broadcom offer AI networking equipment with integrated optical interfaces, while Marvell sells optical DSP technology and has expanded into optical scale-up interconnect technology following its acquisition of Celestial AI. These companies compete at different levels of the same supply chain.

Coherent, Lumentum, InnoLight, Eoptolink and Applied Optoelectronics contribute optical modules, lasers and optical subassemblies for high-speed data-center networks. Coherent operates across silicon photonics, InP and VCSEL technologies, while Lumentum also supplies optical circuit-switching products. Module performance, delivery scale, qualification status and optical power efficiency will influence the relative strength of suppliers as 1.6T deployments expand.

Cisco through Acacia and transport vendors such as Ciena and Nokia serve coherent optical connectivity requirements that become more important as AI facilities connect across campuses and regional data centers. Competitive positioning is likely to depend on the ability to supply complete qualified solutions, retain stable laser and packaging capacity, and offer customers a practical migration path toward higher bandwidth without creating excessive replacement costs.

Recent Developments

• April 2026: Marvell announced its acquisition of Polariton Technologies to strengthen advanced optical modulation technologies for future high-speed AI interconnects.

• March 2026: Coherent demonstrated 1.6T pluggable optical transceivers and next-generation 3.2T technologies at OFC, using several photonics and laser approaches.

• March 2026: Marvell and Lumentum demonstrated an optical circuit-switching system interoperating with multiple optical module classes for AI networking.

• March 2026: An industry consortium founded by AMD, Broadcom, Meta, Microsoft, NVIDIA and OpenAI launched an Optical Compute Interconnect multi-source initiative for scale-up links.

• March 2026: Marvell introduced its COLORZ 1600 coherent optical platform and 2nm DSP technology designed for high-capacity data-center interconnects.

Market Outlook

Optical connectivity is expected to remain an essential part of AI infrastructure as accelerator clusters become larger and are distributed across more computing facilities. Pluggable 800G and 1.6T transceivers will continue to account for substantial purchasing volume, while advanced packaging and higher-speed modules create a path to greater bandwidth and improved energy efficiency.

Co-packaged optics and optical circuit switching are likely to capture a larger share of new AI fabric investment, although their adoption will depend on serviceability, field reliability and software orchestration. Established transceiver vendors should continue to benefit from broad optical upgrades, while integrated switch vendors and optical engine developers compete for new architectures as physical electrical links reach practical constraints.

AI Optical Interconnect Market Scope: