Optical I/O for AI Chips Market Report Scope and Overview:

The Optical I/O for AI Chips Market size was valued at USD 1.08 Billion in 2025 and is projected to reach USD 12.36 Billion by 2035, growing at a CAGR of 27.60% during the forecast period.

The Optical I/O for AI Chips Market is expanding rapidly as large-scale AI training and inference clusters outgrow the bandwidth and power-efficiency limits of traditional copper and pluggable optical interconnects. Hyperscale operators building ever-larger GPU and accelerator clusters are pushing chip-to-chip and rack-to-rack link speeds well beyond what electrical signaling can sustain over meaningful distances, driving a shift toward optics integrated directly at or near the package. Investment is concentrating around co-packaged optics, silicon photonics engines, and heterogeneous integration platforms capable of supporting multi-terabit links at substantially lower power per bit than pluggable transceivers. The partnership between switch ASIC manufacturers, laser source suppliers, and package assembly providers is getting faster as the technology matures from the initial stage to the wider adoption phase.

⮞In August 2025, Celestial AI secured USD 250 million funding in its Series C1 funding round with a USD 2.5 billion valuation for its photonic fabric interconnect technology before AI infrastructure deployment.

U.S. Optical I/O for AI Chips Market Size Outlook

The U.S. Optical I/O for AI Chips Market was valued at approximately USD 0.57 Billion in 2025 and is projected to reach USD 6.64 Billion by 2035, expanding at a CAGR of 27.90% over the forecast period.

The United States demand is influenced by the concentration of hyperscale cloud operators and AI accelerator developers driving early qualification of co-packaged optics and silicon photonics engines for next-generation GPU clusters. The U.S. is a favorite of hyperscalers, cloud operators, and chip design companies. Given the proximity to hyperscaler engineering teams and switch ASIC design centers and persistent venture and strategic capital inflows are fueling a rich ecosystem of specialized optical interconnect startups, the Optical I/O for AI Chips Market is primely suited for first-article qualification. We continue to see investment in capacity expansion across the country driven by increasing focus on supply chain resilience and reshoring of advanced packaging capabilities, with co-packaged optics for AI switches emerging as an area of strong customer interest.

⮞In May 2025, Avicena raised USD 65 Million in a Series B round led by Tiger Global with participation from SK Hynix, supporting development of ultra-low-power microLED-based optical interconnects for AI accelerator links.

By Product Type: LPO Leads, CPO Grows Fastest

The LPO segment dominates the Optical I/O for AI Chips market due to lower integration complexity, cost efficiency, and compatibility with existing switch and transceiver qualification processes. Linear pluggable optics enable hyperscalers and data center operators to deploy higher speed optical links without changing their existing architectures or making significant changes to their infrastructure. An existing supply chain, wide vendor availability, and relatively easy maintenance further support adoption across existing AI infrastructure. As cloud operators continue to grow GPU clusters and upgrade data center networks, LPO remains an attractive solution for near-term deployments where cost, qualification timelines and implementation flexibility drive purchasing decisions.

The CPO segment is the fastest growing segment as hyperscalers and switch ASIC vendors focus on power efficiency and bandwidth density for the next generation AI infrastructure. Co-packaged optics place optical engines closer to switching silicon and reduce the electrical interconnect loss and power consumption associated with conventional pluggable architectures. The growing demand for higher-radix switches, and higher network throughput, underpins investment in optical integration, advanced packaging and manufacturing capabilities. Qualification, thermal management, serviceability and production yields remain key considerations, but continued platform development positions CPO for wider adoption in high-performance AI data center networks.

⮞In February 2026, Ciena introduced its Vesta 200 6.4T CPX optical engine, designed for co-packaged optics and next-generation AI scale-up and scale-out networks, with potential power savings of up to 70%.

By Technology: Silicon Photonics Leads, Heterogeneous Integration Grows Fastest

Silicon Photonics is expected to lead the Optical I/O for AI Chips Market owing to the maturity of the CMOS compatible photonic technologies that allows the semiconductor manufacturers to leverage the existing fabrication infrastructure to enable scalable production. Compatibility with existing semiconductor manufacturing technologies leads to better production efficiency, lower costs, and better integration with leading-edge electronic components. These benefits will reduce unit costs and qualification times, while improving manufacturability, scalability and consistency. Continued demand for high-bandwidth, energy-efficient optical interconnects for AI data centers and hyperscale cloud infrastructure is also driving adoption.

The Heterogeneous/Hybrid Integration segment is experiencing the fastest growth as vendors increasingly combine indium phosphide lasers with silicon photonic waveguides to achieve the optical performance, power efficiency and integration density required for next-generation interconnects. This approach combines the efficient light generation of compound semiconductors with the scalability and integration benefits of silicon photonics. Hybrid optical engines are gaining interest as the increased bandwidth requirements in AI accelerator clusters and high-radix switches are driving demand. Investments in advanced packaging, laser integration, thermal management and manufacturing processes are accelerating platform development and qualification across vendor roadmaps.

⮞In March 2026, HyperLight, UMC, Wavetek and Jabil announced a collaboration to scale thin-film lithium niobate photonics for hyperscale AI data-center interconnects, combining photonic technology with foundry manufacturing and high-volume assembly capabilities.

By Data Rate: 800G Leads, 3.2T and Above Grows Fastest

The 800G segment leads the Optical I/O for AI Chips Market, with this data rate being increasingly incorporated into current-generation AI accelerator and switch platforms for large-scale deployment. The broad ecosystem support from optical engine, connector, transceiver and test equipment vendors improves availability and interoperability of components and strengthens its market position. Established manufacturing capabilities and qualification processes also help hyperscalers and data center operators to manage deployment costs and integration timelines. AI workloads continue to grow and fuel the expansion of GPU clusters and high-speed networking infrastructure, with 800G solutions at the heart of near-term capacity upgrades.

The 3.2T and above segment is growing the fastest as next-generation AI accelerator clusters require substantially higher network bandwidth to support increasingly distributed workloads and large-scale GPU interconnects. Semiconductor companies and makers of optical components are designing solutions that can deliver higher aggregate throughput while managing power consumption and signal integrity, driven by higher-radix switches and more sophisticated networking architectures. Investment is flowing into optical engines, silicon photonics, advanced packaging and high-speed test infrastructure as early platforms are developed and design wins are achieved.

⮞In September 2025, Ciena, HyperLight and McGill University demonstrated 448 Gbps-per-lane PAM4 optical transmission over 500 metres of fiber, supporting development of next-generation 3.2T optical interconnects for data-center connectivity.

By Application: AI Training Leads, AI Inference Grows Fastest

The AI Training Segment is the leading segment in the Optical I/O for AI Chips Market due to the heavy bandwidth requirements of large multi-node training clusters that rely on dense low-latency chip-to-chip and rack-to-rack optical links. Training ever larger AI models requires high-throughput communication between GPUs and accelerators, driving the need for more sophisticated optical interconnect solutions. The capital intensity of these clusters allows hyperscalers to optimize for performance, energy efficiency and scalability in the deployment of next generation optical technologies.

The AI Inference segment is the fastest growing segment as inference workloads are being deployed in distributed, multi-accelerator environments that require high-bandwidth, low-latency interconnects. Serving large language models at scale requires constant communication across accelerators, especially for applications that require high throughput and low latency. As enterprises and cloud operators build out AI inference infrastructure, the demand for efficient optical connectivity is increasing. This change moves optical I/O beyond deployments focused on training and allows for the expansion of the market across a variety of AI workloads.

By End User: Hyperscalers and Cloud Providers Lead, GPU Cloud Providers Grow Fastest

The Hyperscalers and Cloud Providers segment of the Optical I/O for AI Chips Market is expected to dominate due to the large capital capacity to deploy advanced interconnect technologies and initiate early qualification programs with switch ASIC and optical component vendors. These companies run big AI data centers that need high bandwidth, low latency, and more power efficiency. Their relationships direct with semiconductor vendors and networking providers also allow for customized integration, early technology adoption and large volume procurement, fueling demand for optical I/O solutions across next-generation AI infrastructure deployments.

GPU Cloud Providers is the fastest growing segment, with specialized AI infrastructure operators growing fast to meet rising demand for accelerator capacity. These providers are increasingly leveraging high-performance GPU clusters to fuel model training, inference, and other compute-intensive workloads. The development of advanced optical interconnects results in enhanced communication efficiency between accelerators and scalable cluster architectures. GPU cloud providers are vying for enterprise customers and AI developers with investments in higher bandwidth connectivity and energy-efficient networking. Next-gen optical I/O technologies are being adopted by AI-centric cloud platforms.

⮞In March 2025, NVIDIA launched the Blackwell Ultra AI platform, as well as updates to infrastructure and software to help GPU cloud providers and other operators scale their AI training and test-time inference workloads.

Regional Analysis

North America Optical I/O for AI Chips Market Insights

North America accounted for the highest revenue share of around 56.80% in the Optical I/O for AI Chips Market in 2025. This is attributed to the presence of hyperscale cloud operators, AI accelerator designers, and switch ASIC vendors in the region, promoting early qualification of co-packaged optics and silicon photonics engines. A dense cluster of optical interconnect startups in the region, many backed by strategic investment from leading chip and cloud companies, continues to anchor global innovation in the space, with new design centers and manufacturing partnerships announced throughout 2025 and 2026 reinforcing the region’s lead.

The United States dominated regional revenue, driven by the concentration of hyperscale data center builds and AI accelerator design work across both established semiconductor firms and niche optical interconnect startups. Canada continued to be a smaller but steady contributor, with some photonics research and component supply activity. Continued venture capital inflows across the broader North American optical interconnect ecosystem continued to support new entrants pursuing co-packaged optics and hybrid integration platforms.

⮞In October 2025, Broadcom reported one million cumulative 400G-equivalent port device-hours of flap-free co-packaged optics operation at Meta, providing evidence of reliability for CPO deployment in hyperscale AI infrastructure.

Europe Optical I/O for AI Chips Market Insights

Europe held a meaningful share of global revenue, supported by an established base of photonics research institutions and component manufacturers, along with growing investment in sovereign AI infrastructure that is increasingly incorporating advanced optical interconnect technology into new data center builds. National research programs across several countries continue funding advanced photonic integration research that feeds directly into commercial component development.

The United Kingdom continued to lead regional demand at roughly 26.70% of European revenue, supported by a concentration of photonics component and packaging specialists, while Germany and France continued contributing meaningful research and manufacturing capacity tied to the broader European semiconductor ecosystem. Growing coordination between national research institutes and commercial photonics manufacturers continues strengthening the region's position in advanced packaging.

⮞In September 2025, STMicroelectronics announced the STARLight European consortium, involving 24 companies and universities across 11 EU countries to develop 300 mm silicon photonics manufacturing and optical modules for data centers and AI clusters.

Asia Pacific Optical I/O for AI Chips Market Insights

Asia Pacific is advancing at the fastest pace, growing at a CAGR of approximately 33.50% from 2026 to 2035, driven by expanding optical component manufacturing capacity, rising regional AI infrastructure investment, and a deep base of contract manufacturers supplying global switch and transceiver vendors. Government-backed semiconductor and photonics investment programs across the region continue lowering the barrier for new manufacturing capacity dedicated to optical interconnect components.

China held a substantial share of regional revenue, supported by an expanding base of optical component and module manufacturers serving both domestic and export AI infrastructure demand, while Taiwan and South Korea continued contributing significant advanced packaging and laser source manufacturing capacity. Japan continued contributing specialized component manufacturing tied to its established optical communications supply chain.

⮞In July 2026, UMC and SILITH Technology announced the first mass-production delivery of silicon photonics wafers from UMC’s Singapore facility, supporting high-volume manufacturing of 1.6T photonic integrated circuits for AI data-center networks.

MEA and Latin America Optical I/O for AI Chips Market Insights

The Middle East and Africa and Latin America both showed early-stage growth, driven by new sovereign AI infrastructure investment programs and gradually increasing interest from global hyperscale operators in diversifying data center locations beyond traditional hubs. Government-backed digital infrastructure initiatives across both regions continue to lower the barrier for new data center investment incorporating advanced interconnect technology from the outset.

Israel led Middle East and Africa demand, supported by the country's concentration of photonics and semiconductor design talent, while Brazil led Latin America, benefiting from growing regional cloud infrastructure investment tied to expanding AI service demand. The United Arab Emirates and Mexico continued contributing secondary capacity within their respective regions as sovereign AI investment programs matured.

⮞In February 2026, Tower Semiconductor announced that it was scaling its silicon photonics platform for 1.6T optical modules designed for NVIDIA networking protocols, strengthening Israel-linked manufacturing capabilities for high-bandwidth AI data-center connectivity.

Market Dynamics

Growth Drivers: Rising AI Cluster Bandwidth Demand and Power Efficiency Requirements

The industry is driven by the escalating bandwidth and power-efficiency requirements of large-scale AI training and inference clusters, which have outgrown the practical limits of copper and pluggable optical interconnects at the data rates now required for competitive cluster performance. Operators building next-generation GPU and accelerator clusters increasingly treat interconnect bandwidth as a primary design constraint rather than a secondary consideration, pulling investment directly toward optical I/O technology.

Rising capital investment from hyperscale operators and chip designers in co-packaged optics and silicon photonics continues supporting capacity build-out, as these companies increasingly view optical interconnect performance as a direct driver of overall AI cluster throughput and cost per computation. Strategic investment from chip and cloud companies into specialized optical interconnect startups continues accelerating the pace at which new technology moves from demonstration to qualified production.

⮞In March 2026, AMD, Broadcom, Meta, Microsoft, NVIDIA and OpenAI established the Optical Compute Interconnect Multi-Source Agreement to promote an open, multi-vendor specification for optical scale-up interconnects in AI infrastructure.

Restraints: Integration Complexity and Reliability Qualification

Integration complexity between optical engines and switch or accelerator packages continues posing a restraint, with thermal management, yield, and repairability challenges at the package level requiring extensive qualification before volume deployment can proceed. These challenges are particularly acute for co-packaged designs, where a single component failure can require replacement of the entire package rather than a simple pluggable module swap.

Reliability qualification timelines for co-packaged and hybrid-integrated optical components continue posing execution risk, particularly given the higher cost of field failure at the package level compared with pluggable modules that can be swapped without disrupting the broader system. Hyperscale operators continue to require extensive field trial data before committing to large-scale deployment of newer co-packaged platforms.

Opportunities: Inference Workload Expansion and Emerging Market Infrastructure Build-Out

Rising adoption of optical interconnects for distributed inference workloads represents a significant opportunity, as inference deployments increasingly require the same high-bandwidth, low-latency links previously associated mainly with training clusters. Vendors able to offer interconnect solutions optimized for both training and inference workloads are positioned to capture a broader share of overall AI infrastructure spending.

Expanding sovereign and regional AI infrastructure investment across emerging markets offers a second substantial opportunity, as governments and regional cloud operators increasingly incorporate advanced optical interconnect technology into new data center build-outs from the outset. This early-stage incorporation reduces the retrofit burden that has historically slowed adoption of new interconnect technology in established data center markets.

⮞In March 2026, Marvell and Mojo Vision announced a collaboration to develop high-density micro-LED optical interconnects for AI data centers, targeting bandwidth density, power efficiency and connectivity between accelerator arrays and shared memory resources.

Optical I/O for AI Chips Companies are:

• Marvell Technology, Inc.

• Coherent Corp.

• Lumentum Holdings Inc.

• Broadcom Inc.

• Ayar Labs, Inc.

• Avicena Tech Corp.

• Ranovus Inc.

• POET Technologies, Inc.

• Cisco Systems, Inc.

• Intel Corporation

• NVIDIA Corporation

• Innolight (Shanghai) Technology Co., Ltd.

• Source Photonics, Inc.

• Teramount Ltd.

Optical I/O for AI Chips Market Report Scope: