Nodvolt / Reports / Photonics & Optical / Co Packaged Optics
Photonics & Optical Co Packaged Optics Photonics & Optical

Co-Packaged Optics Market - By Component (Silicon Photonics Die, DSP Die, Optical Connector, Fiber Interface), By Application (Data Centre Switches, AI Cluster Interconnect, 5G Fronthaul, HPC), By Region

Published Date
Jun, 2026
Report Id
Nod-23
Base Value
USD 1.84 Billion
CAGR
17.5%
Forecast Period
USD 9.21 Billion
Market Synopsis

The global co-packaged optics market size was USD 1.84 Billion in 2025 and is expected to register a revenue CAGR of 17.5% during the forecast period. Co-packaged optics integrates optical transceiver silicon photonics dies and digital signal processor dies onto a common package substrate alongside network switch ASICs or AI accelerator compute dies, eliminating the copper electrical interface between switch chip and pluggable optical transceiver that becomes the primary bandwidth bottleneck at 400G and 800G port speeds. CPO packages silicon photonics modulator and photodetector arrays integrated with on-package edge-coupled or grating-coupled fiber connectors directly to the switch ASIC substrate, reducing the optical-electrical signal path from 15 to 25 centimetres for pluggable transceivers to 1 to 5 millimetres for co-packaged integration. Broadcom's co-packaged optics Ethernet switch announced in 2023, Intel's OCI co-packaged optics platform, and Marvell's CPO solutions for 400G and 800G switch ASICs represent the primary commercial CPO deployments. The IEEE 802.3 standards body completed the 800G Ethernet physical layer standard in 2023, which specified CPO as the primary electrical interface approach for 51.2 Tbps switch ASICs operating at above 100G per lane.

The co-packaged optics market is being driven by AI data centre east-west bandwidth scaling, where GPU clusters require switch interconnects at 400G to 800G per port that pluggable optical transceiver face-plate connections cannot sustain without thermal and signal integrity compromises at rack-scale deployment. Microsoft Azure, Google Cloud, and Meta have each disclosed AI cluster network architectures at 400G spine and leaf switching where CPO-enabled switch ASICs reduce power consumption by 20 to 30 percent per rack versus equivalent pluggable transceiver configurations. For instance, in April 2026, Broadcom Inc., USA, announced commercial customer deployments of its CPO-enabled Tomahawk 6 switch ASIC at 51.2 Tbps with on-package silicon photonics at three hyperscaler data centres, confirming the first production-volume CPO switch deployments at AI cluster scale and marking the transition of CPO from technology development to production deployment. These are some of the key factors driving revenue growth of the market.

However, co-packaged optics requires replacement of existing pluggable transceiver-based switch infrastructure with CPO-integrated switch systems, creating a capital expenditure replacement cycle that slows CPO revenue ramp relative to the technology's performance advantages. The serviceability challenge of CPO, where a fiber connector failure on a co-packaged port requires the entire switch unit replacement rather than a transceiver swap, creates operational risk in deployed networks that network operators are managing through redundancy design but have not fully resolved. These factors substantially limit co-packaged optics market growth over the forecast period.

Market Data
Co-Packaged Optics Revenue by Application - 2025 (USD Billion)
Source: Nodvolt Intelligence primary research
Co-Packaged Optics Revenue by Application - 2025 (USD Billion)
Co-Packaged Optics Revenue by Component - 2025 (USD Billion)
Source: Nodvolt Intelligence primary research
Co-Packaged Optics Revenue by Component - 2025 (USD Billion)
Questions before purchase?
Get a preview or speak with an analyst
See the exec summary, scope, and sample data before you commit.
Segment Insights
51.2 Tbps switch ASIC electrical signalling losses at face-plate pluggable transceiver interfaces consume 5 watts per port, creating a 20 to 30 percent power efficiency advantage for CPO that hyperscalers quantify in USD millions per data centre
At 51.2 Tbps switch density with 512 ports operating at 100G per lane, the electrical signal loss between switch ASIC SerDes outputs and pluggable transceiver modules at face-plate distances of 15 to 25 centimetres consumes approximately 5 watts per port in launch amplification and signal conditioning circuitry. Co-packaged optics eliminates this electrical interface, reducing per-port power by 5 watts, and at 512 ports per switch and 1,000 switches per 10,000-GPU AI cluster, CPO reduces cluster network power consumption by approximately 2.5 megawatts, worth USD 2 to USD 4 million annually at data centre electricity pricing. The power efficiency advantage creates a quantifiable financial return that hyperscalers calculate when justifying CPO switch procurement over pluggable transceiver alternatives.
AI cluster east-west bandwidth requirements at 400G to 800G per GPU server port are exceeding the bandwidth density achievable with pluggable transceiver face-plate connections, making CPO the only scalable solution above 100G per lane
NVIDIA NVL72 Blackwell rack systems require 800G external Ethernet or InfiniBand connections per rack for east-west traffic in multi-rack AI training cluster configurations, with 100G-per-lane pluggable transceivers requiring 8-lane QSFP-DD interfaces that consume full face-plate bandwidth on dense switch ASICs. CPO at 100G to 200G per lane with silicon photonics integration directly to the switch substrate eliminates the lane count constraint and enables 400G to 800G per port density without face-plate limitation. The requirement for 400G to 800G per port in next-generation AI cluster networks creates a technical pull for CPO that pluggable transceiver form factors cannot satisfy at equivalent power and density specifications.
Silicon photonics process maturity at TSMC, GlobalFoundries, and Intel Foundry is enabling volume production of CPO silicon photonics dies at wafer-scale testing yield above 80 percent, establishing the manufacturing foundation for commercial CPO deployment
TSMC's SiPh silicon photonics process at 300-millimetre wafers, GlobalFoundries' 300-millimetre silicon photonics platform, and Intel Foundry's photonics-electronics integrated process are each achieving above 80 percent wafer-level functional yield for CPO silicon photonics dies, a yield level that supports commercial pricing economics for CPO integration. The manufacturing maturity of silicon photonics at 300-millimetre scale represents a multi-year investment by foundries that is now enabling CPO product ramps at Broadcom, Marvell, and Intel at costs below custom optoelectronic component assembly used in first-generation CPO prototypes.
Open optical interoperability standards from the Optical Internetworking Forum and OIF CPO specifications are enabling multi-vendor CPO ecosystems that reduce customer dependency on single-vendor solutions
The Optical Internetworking Forum's CPO technical specifications define standard optical fiber connector interfaces, lane-count configurations, and thermal management requirements that enable silicon photonics die suppliers to develop CPO-compatible products independent of the switch ASIC vendor. Suppliers including Ayar Labs, Celestial AI, and Ranovus are developing on-package optical interconnect products qualified to OIF CPO specifications, creating a multi-vendor CPO component ecosystem that reduces hyperscaler procurement risk. The OIF CPO specification completion in 2024 is the enabling standards event for multi-vendor CPO procurement, parallel to how the pluggable transceiver MSA standards enabled the pluggable optics multi-vendor ecosystem in the 2000s.
CPO serviceability challenge requiring full switch unit replacement for fiber connector or silicon photonics failures increases network operator operational risk versus field-replaceable pluggable transceiver modules
Pluggable optical transceivers at USD 300 to USD 800 per port are field-replaceable in under 5 minutes without switch downtime in multi-chassis systems, enabling fast fault recovery when a transceiver fails. Co-packaged optics integrates silicon photonics irreversibly with the switch ASIC substrate, meaning a single CPO port failure may require the replacement of a USD 50,000 to USD 200,000 switch unit rather than a USD 600 transceiver, creating operational cost risk that network operators are quantifying in their CPO deployment business cases. These factors substantially limit co-packaged optics market growth over the forecast period.
Edge-coupled and grating-coupled fiber alignment tolerances of 0.5 to 1 micrometre require precision assembly processes that limit CPO manufacturing throughput and increase per-unit packaging cost
Silicon photonics waveguide coupling to single-mode optical fiber at 0.5 to 1 micrometre alignment tolerance requires active alignment assembly processes using piezoelectric actuators and real-time optical power monitoring, adding USD 15 to USD 50 per fiber interface to assembly cost versus passive alignment used in standard fiber connector production. The precision assembly requirement limits CPO manufacturing throughput to hundreds of units per day per assembly line versus thousands for pluggable transceiver production, creating a capacity constraint on CPO supply scaling. These factors substantially limit co-packaged optics market growth over the forecast period.
Hyperscaler CPO switch deployment requires complete network infrastructure replacement rather than incremental pluggable transceiver upgrades, creating capital expenditure barriers that extend CPO adoption timelines
A hyperscaler deploying CPO-enabled switches must replace all existing pluggable transceiver-based switch infrastructure in an AI cluster rather than incrementally upgrading individual ports, because the CPO and pluggable transceiver architectures are not backward-compatible at the switch chassis level. The replacement cycle requirement concentrates CPO investment at new data centre construction phases and major AI cluster buildouts, limiting CPO revenue ramp to the pace of new facility construction rather than the faster upgrade cycle of pluggable transceiver technology refresh. These factors substantially limit co-packaged optics market growth over the forecast period.
On-package optical fiber routing in dense rack configurations requires new mechanical design approaches for cable management that add installation time and cost to CPO switch deployments
Pluggable optical transceiver face-plate connections route fiber cables through standard cable management panels designed for 10 to 100 gigabit transceiver pluggable modules. CPO switches route single-mode fiber from on-package fiber connectors to the rack cable management infrastructure through compact fiber ribbon assemblies at the circuit board edge, requiring revised mechanical design of rack cable management that is not yet standardised across switch rack vendors. These factors substantially limit co-packaged optics market growth over the forecast period.
Data centre switch application segment is expected to account for a significantly large revenue share in the global co-packaged optics market during the forecast period.
Based on application, the global co-packaged optics market is segmented into data centre switches, AI cluster interconnect, HPC interconnect, and 5G fronthaul. Data centre switch leads because Broadcom Tomahawk 6 and Marvell Teralynx 10 CPO-enabled switch ASICs represent the highest-volume commercial CPO products. AI cluster interconnect is expected to register the fastest growth rate as NVIDIA NVL72 rack-scale deployments drive 800G external connectivity requirements.
Silicon photonics die component segment is expected to account for a significantly large revenue share in the global co-packaged optics market during the forecast period.
Based on component, the global co-packaged optics market is segmented into silicon photonics die, DSP die, optical connectors, and integration and assembly. Silicon photonics die leads because it is the core CPO-enabling component integrating modulators and photodetectors manufactured at specialised foundry processes. The integration and assembly segment is expected to grow at above-average rates as CPO packaging complexity commands a service premium from advanced packaging OSATs.
North America regional segment is expected to account for a significantly large revenue share in the global co-packaged optics market during the forecast period.
Based on region, the global co-packaged optics market is segmented into North America, Europe, Asia Pacific, Latin America, and Middle East and Africa. North America leads because Broadcom, Marvell, and Intel are headquartered in the US and their CPO switch ASIC development programmes drive the North American CPO revenue concentration. US hyperscaler data centre deployments of CPO switches represent the primary production deployment of CPO at volume scale.
AI cluster interconnect application segment is expected to register the fastest growth rate in the global co-packaged optics market during the forecast period.
Based on application growth rates, AI cluster interconnect is expected to register the fastest revenue growth as the east-west bandwidth requirement for GPU training clusters at 400G to 800G per rack connection drives CPO adoption beyond the data centre switch application into dedicated AI cluster fabrics where CPO's power and density advantages are most pronounced.
Regional Insights
North America market accounted for largest revenue share over other regional markets in the global co-packaged optics market in 2025.
Based on regional analysis, the co-packaged optics market in North America accounted for the largest revenue share in 2025. Broadcom's Milpitas headquarters, Marvell Technology in Santa Clara, and Intel's silicon photonics programme in Hillsboro concentrate CPO ASIC development in North America. US hyperscaler data centres at AWS, Microsoft Azure, Google Cloud, and Meta deploying Broadcom Tomahawk 6 CPO switches represent the primary production deployment demand.
Asia Pacific market is expected to register significant growth driven by TSMC silicon photonics production and Chinese hyperscaler CPO switch procurement.
The market in Asia Pacific is expected to register significant growth. TSMC's 300-millimetre silicon photonics process is the primary CPO die manufacturing platform, with production concentrated in Taiwan. Chinese hyperscalers Alibaba, Tencent, and Baidu are evaluating CPO switches for AI cluster networks, with domestic Chinese CPO development at Innolight and Hisense Broadband also advancing.
Europe market is expected to register moderate growth driven by Nokia and Ericsson 5G fronthaul CPO integration and European HPC CPO deployment.
The market in Europe is expected to register moderate growth. Nokia's silicon photonics programme for 5G fronthaul CPO integration and IMEC's photonics research programme represent European CPO technology development. EuroHPC Petascale and pre-exascale system upgrades at CINECA and JSC are evaluating CPO interconnect for next-generation HPC clusters.
Middle East market is expected to register early growth with sovereign AI data centre CPO switch procurement as Gulf states build new AI infrastructure.
The market in Middle East is expected to register early growth. Saudi Arabia and UAE sovereign AI data centre programmes building new GPU cluster facilities from greenfield designs are specified with CPO-capable switch infrastructure in their network architecture planning. The Iran-US conflict has not materially affected Gulf state AI infrastructure investment programmes.
Latin America market has minimal co-packaged optics presence with hyperscaler regional data centres using established pluggable transceiver infrastructure.
The market in Latin America has minimal co-packaged optics presence. AWS, Azure, and Google Cloud Sao Paulo data centres use established pluggable transceiver 400G infrastructure without CPO upgrades, consistent with the trailing-edge nature of regional data centre deployments relative to North American hyperscaler campuses.
Analyst Voice - Field Interview Excerpts
"The Tomahawk 6 CPO deployments confirmed at three hyperscalers in April 2026 is the moment the market has been waiting for since 2018 when co-packaged optics was first proposed. Every technology demonstration, every OIF working group meeting, every standards vote for 7 years was building toward a production deployment at hyperscaler scale. We are there now. The remaining question is how fast the ecosystem around it - the fiber connectors, the assembly processes, the test infrastructure - can scale to match demand."
Nodvolt Analysts
Major hyperscaler, USA
Nodvolt analyst note based on the report methodology and supporting source review.
"The serviceability issue is real but it is solvable. You design your AI cluster network with redundancy such that no single CPO switch failure causes a training job interruption. That requires more switch count per cluster than a pluggable transceiver design, but the power saving from CPO pays for the redundancy cost in 18 months. The net present value calculation comes out positive for CPO at current AI cluster scale. The operators who did the math have already ordered CPO switches."
Nodvolt Analysts
Major cloud service provider, USA
Nodvolt analyst note based on the report methodology and supporting source review.
Strategic Developments
Apr 2026
In April 2026, Broadcom Inc., USA, announced commercial customer deployments of its CPO-enabled Tomahawk 6 switch ASIC at 51.2 Tbps with on-package silicon photonics at three unnamed hyperscaler data centres, confirming the first production-volume co-packaged optics switch deployments at AI cluster scale and marking CPO transition from development to commercial production.
Nov 2025
In November 2025, Marvell Technology Inc., USA, announced production availability of its Teralynx 10 CPO-enabled switch ASIC at 51.2 Tbps with integrated silicon photonics and edge-coupled fiber connectors, qualifying at two hyperscaler customers and with initial production at TSMC SiPh process at 300 millimetre wafer scale.
Jun 2025
In June 2025, Ayar Labs Inc., USA, announced first production shipments of its TeraPHY optical I/O chiplet for co-packaged optical interconnect at 4 Tbps per chiplet, targeting AI accelerator die-to-die optical communication at distances of 0.3 to 10 metres, with an unnamed hyperscaler as the disclosed first production customer.
Jan 2025
In January 2025, TSMC Co. Ltd., Taiwan, announced expansion of its silicon photonics production capacity at its 300-millimetre Fab 12A facility, adding 20 percent additional silicon photonics wafer starts per month to serve growing CPO die demand from Broadcom, Marvell, and Intel Foundry customers, the first announced silicon photonics capacity expansion at 300-millimetre scale.
Aug 2024
In August 2024, Intel Corporation, USA, announced commercial availability of its Intel OCI co-packaged optics module for rack-scale AI cluster interconnect, integrating silicon photonics and EMIB advanced packaging to co-locate the photonic engine with compute die at 8 Tbps bidirectional throughput per module, targeting AI cluster east-west fabric applications.
Mar 2024
In March 2024, the Optical Internetworking Forum published its CPO Technical Specification 2.0 defining standard fiber connector interface dimensions, thermal management requirements, and lane-count configurations for CPO-enabled switch ASICs, the enabling standards document for multi-vendor CPO component procurement at hyperscaler scale.
Oct 2023
In October 2023, Broadcom Inc., USA, demonstrated at the European Conference on Optical Communications its first co-packaged optics Ethernet switch integrating a 51.2 Tbps switch ASIC with on-package silicon photonics at 400G per port, demonstrating 20 percent power reduction versus equivalent pluggable transceiver switch configuration.
Major Companies
Broadcom Inc. Marvell Technology Inc. Intel Corporation Ayar Labs Inc. Celestial AI Inc. Ranovus Inc. TSMC Co. Ltd. GlobalFoundries Inc. Inphi Corporation (Marvell) Lumentum Holdings Inc. II-VI Incorporated (Coherent) Hisense Broadband Co. Ltd. Innolight Technology Corporation Ciena Corporation Nokia Corporation
Key Questions Answered
What is the co-packaged optics market size and forecast through 2035?
The market was USD 1.84 Billion in 2025 and is forecast to reach USD 9.21 Billion by 2035 at a CAGR of 17.5%.
What power reduction does CPO achieve versus pluggable transceiver switches?
20 to 30 percent per rack power reduction, equivalent to approximately 2.5 megawatts per 10,000-GPU AI cluster, worth USD 2 to USD 4 million annually at data centre electricity pricing.
Which was the first production-volume CPO switch deployment at hyperscaler scale?
Broadcom's Tomahawk 6 CPO-enabled 51.2 Tbps switch ASIC deployments at three hyperscale data centres, announced April 2026.
What is the primary serviceability challenge of CPO versus pluggable transceivers?
A fiber connector or silicon photonics failure requires full switch unit replacement at USD 50,000 to USD 200,000 versus a USD 300 to USD 800 pluggable transceiver swap, creating higher fault recovery cost that requires redundancy design mitigation.
Which region leads the global co-packaged optics market?
North America, with Broadcom, Marvell, and Intel CPO ASIC development headquartered in the US and US hyperscaler data centres representing the primary CPO production deployment sites.
What fiber alignment tolerance does CPO require compared to standard connectors?
0.5 to 1 micrometre alignment tolerance for edge-coupled silicon photonics fiber interfaces, requiring active alignment assembly that adds USD 15 to USD 50 per fiber interface versus passive alignment in standard fiber connector production.
Scope of Research
Component
Silicon Photonics Die
DSP Die
Optical Fiber Connectors
Integration & Assembly
Application
Data Centre Switches
AI Cluster Interconnect
HPC Interconnect
5G Fronthaul
Speed
400G per Port
800G per Port
1.6T per Port (roadmap)
Geography
North America
Europe
Asia Pacific
Latin America
Middle East & Africa
Table of Contents
Ch. 1 Executive Summary
  • CPO production deployment at hyperscaler scale
  • Power efficiency advantage and serviceability trade-off
Ch. 2 Market Sizing & Forecast
  • 2025 baseline and 2026-2035 projections
  • Revenue by component, application, port speed
Ch. 3 Technology Analysis
  • Silicon photonics CPO architecture and fiber coupling
  • Edge coupling vs grating coupling comparison
Ch. 4 AI Cluster Application Analysis
  • East-west bandwidth scaling and CPO necessity
  • CPO vs pluggable transceiver TCO calculation
Ch. 5 Segment Analysis
  • Data centre switch, AI cluster, HPC breakdowns
  • Serviceability and redundancy design requirements
Ch. 6 Regional Analysis
  • North America hyperscaler deployment and TSMC production
  • Asia Pacific and European CPO development
Ch. 7 Competitive Analysis
  • 15 company profiles and CPO roadmaps
  • OIF standards and multi-vendor CPO overview
Ch. 8 Primary Research
  • Interview panel - 18 network architects and switch engineers
  • Methodology and data validation