Co-Packaged Optics Market 2026

Co-packaged optics (CPO) places optical engines beside a switch or compute device within a shared package, shortening high-speed electrical paths before data enters fibre. Its proposition is improved interconnect power efficiency and bandwidth density. This report treats CPO as data-movement technology, not optical execution of AI models; silicon-photonic pluggable transceivers are not automatically CPO.

The commercial scope spans optical engines, electrical-interface circuits, laser sources, fibre connectivity, package assembly and testing, alongside complete CPO systems. Component capability, manufacturing-platform availability and finished-system readiness are different milestones. The central competitive question is whether a qualified architecture reduces the cost and operational risk of moving data—not simply whether an individual device delivers an attractive laboratory result.

Commercialization is product-specific. NVIDIA reported Spectrum-X Ethernet Photonics in production on May 31, 2026. Broadcom’s October 8, 2025 Davisson announcement used shipping language while its availability section specified early-access sampling. Neither statement establishes widespread fleet deployment. STMicroelectronics’ March 9, 2026 announcement similarly distinguishes high-volume PIC100 silicon-photonics production from its planned TSV platform supporting future near-packaged optics and CPO.

Packaging and operability are becoming competitive differentiators. NVIDIA describes screened optical engines, detachable fibre connections and automated final fibre attachment. Broadcom reported flap-free testing at Meta in a high-temperature laboratory environment in 2025. These developments address manufacturing and qualification risks, but laboratory link stability does not establish lifetime system reliability or repair economics.

Standards provide building blocks, not complete interchangeability. OIF has implementation agreements for CPO modules and external pluggable laser sources. Its CPO module agreement nevertheless leaves optical-connector choices to implementers. Buyers therefore need to distinguish standards-based interfaces from demonstrated multivendor compatibility across an entire switch.

AI increases the commercial value of efficient, reliable data movement. NVIDIA’s technical discussion links low-jitter fabric behavior to training and inference efficiency and to expert dispatch in mixture-of-experts models. This supports demand for better interconnects, but improvements from routing, topology and protocol optimization cannot all be attributed to the optical package.

The cumulative effect extends beyond switch ports toward compute-adjacent optical I/O and manufacturing commitments. Ayar Labs and Alchip’s September 2025 partnership targets multi-rack scale-up connectivity through optical I/O and advanced packaging. NVIDIA and Coherent’s March 2026 agreement illustrates the parallel importance of securing optics capacity, development resources and future supply access.

Our synthesis is that AI changes CPO purchasing from a component-efficiency decision into a system-utilization decision. Evaluation should track completed training and inference work, including network energy, cooling and interruptions. Lower optical-link consumption should not be assumed to produce an identical reduction in total facility electricity use.

North America combines visible switch commercialization with investment in optical supply. The announced NVIDIA–Coherent collaboration supports a U.S. manufacturing build-out. The regional commercial priority is operator-led qualification and dependable supply, rather than assuming that U.S. product announcements establish adoption throughout North America.

Europe offers complementary routes through industrial photonics and shared packaging infrastructure. STMicroelectronics’ production platform and the APECS pilot line illustrate manufacturing capability and technology-transfer support. For suppliers, the practical opportunity is to turn photonic integration, testing and reliability capabilities into qualified components and packages—not to equate broad photonics activity with CPO shipments.

Asia-Pacific is directly connected to industrialization through packaging and foundry roadmaps. TSMC’s April 2026 statement targeted on-substrate CPO production during 2026; this review did not independently establish fulfillment of that target. GlobalFoundries’ Singapore operations provide another manufacturing pathway linked to a CPO roadmap.

For Latin America, the inspected primary material did not establish a local CPO deployment or production program. Imported-system qualification and support are sensible opportunities to investigate, but realized regional demand remains unestablished in this review. In the Middle East, prospective AI infrastructure should likewise not be counted as CPO adoption without named architectures, orders and acceptance evidence; an operating regional deployment was not established by the reviewed sources.

Africa has a documented engineering connection: GlobalFoundries acquired Cairo-based InfiniLink to strengthen optical-connectivity design and its optics roadmap. This demonstrates participation in the enabling technology chain, not local volume CPO manufacturing or an installed deployment base.

Within ASEAN, member Singapore provides a specific foundry example through GlobalFoundries’ acquisition of Advanced Micro Foundry and its photonics development plans. This supports location-specific collaboration opportunities; it should not be generalized into bloc-wide CPO deployment.

The European Union has a more explicit shared-infrastructure pathway. APECS supports advanced packaging and qualification, while PIXEurope connects photonic design, fabrication, integration, packaging and testing through a distributed pilot-line model. These initiatives can lower development barriers, but public support for photonics is not proof of CPO purchasing demand.

The G7 semiconductor coordination meeting of September 25–26, 2025 addressed research collaboration, manufacturing and supply-chain resilience. This is relevant context for CPO sourcing and qualification, not evidence of a CPO-specific procurement mandate or technical standard.

For BRICS, the reviewed primary evidence did not establish a common CPO commercialization program. National supplier capabilities and purchasing conditions should be assessed separately rather than treated as one adoption pathway.

For the GCC, no bloc-specific CPO procurement program was established; infrastructure investment potential must be separated from documented switch orders. For NATO, no alliance-level CPO procurement or interface program was established. Supply assurance and security may be useful procurement criteria, but they are not adoption evidence.

North America — United States, Canada and Mexico. In the United States, the NVIDIA–Coherent agreement documents supply commitments, development activity and support for future U.S. optics manufacturing; it does not certify completed CPO capacity. Canada’s CMC Microsystems offers fibre-coupled photonic packaging, electrical connections and thermal-management services relevant to prototype qualification. For Mexico, the reviewed material did not establish a country-specific CPO installation or manufacturing program.

Latin America — Brazil. No inspected primary source documented a Brazil-specific CPO deployment or production program. Candidate integration and maintenance opportunities should therefore be presented as possibilities requiring customer evidence, not as an established domestic market.

Europe — France, Germany, Italy, Spain, United Kingdom and Russia. France’s CEA-Leti and Germany’s Fraunhofer participate in APECS, providing a shared packaging-development route. Italy participates in PIXEurope; Spain has a coordination role through ICFO in Barcelona. The United Kingdom contributes through Southampton’s CORNERSTONE Photonics Innovation Centre. These are development and infrastructure connections, not country-wide deployment measures. For Russia, current CPO production or operating installations were not established by the inspected evidence.

Asia-Pacific — China, Japan and South Korea. China has a documented early commercialization effort: the Broadcom–Tencent–Ruijie collaboration announced in 2022 covered system design, manufacturing and testing roles, with a planned Shenzhen demonstration. It does not measure the current installed base. Japan’s NTT published CPO engine and compact-connector prototype work in January 2024, demonstrating technical depth rather than current fleet adoption. In South Korea, Samsung exhibited a CPO package mock-up at its August 2026 packaging event; a mock-up should not be treated as a shipping system.

Asia-Pacific — India and Australia. IIT Madras’s May 2025 release documents a fibre-array attachment tool and silicon-photonics design, testing and packaging partnerships, indicating enabling capability rather than CPO deployment. CSIRO’s September 2, 2026 announcement that Australia’s Q-APP facility is open documents domestic photonic-chip packaging and testing access, not a high-volume CPO production line.

Stage procurement around operating evidence. Compare CPO, pluggable and near-packaged alternatives using representative workloads. Require a clear distinction between samples, production-qualified products and accepted customer installations. Include power, useful throughput, interruptions and recovery behavior in acceptance criteria.

Design serviceability before committing to integration. Evaluate external-laser replacement separately from optical-engine repair. Test fibre access, connector cleanliness, diagnostic coverage and the consequence of a failed engine for the surrounding switch. Replaceable laser modules do not make every package component field-serviceable.

Qualify the manufacturing chain, not just the photonic die. Assess pre-assembly screening, fibre-attachment processes, thermal margins, package-level testing and failure traceability. Validate the actual connector and management-interface combination rather than assuming standards compliance guarantees interchangeability.

Contract for maturity and continuity. Align capacity reservations with qualification gates, warranty responsibilities and repair arrangements. Build a supplier-location map covering fabrication, lasers, assembly and testing. In geographies with weak deployment evidence, start with narrowly scoped integration pilots instead of assuming broad local demand. These are editorial recommendations, not commitments endorsed by the cited suppliers.

This qualitative desk research uses a reporting cutoff of October 5, 2026. Inspected material includes company announcements and technical articles, OIF documents, public research-institution pages and European Commission publications. Dated documents provide commercial milestones and historical context; undated service pages establish only the capabilities they list.

Evidence was classified as an announcement, sampling statement, vendor-reported production milestone, laboratory qualification result, prototype or enabling infrastructure. These categories were not treated as interchangeable. In particular, laboratory link testing was not extrapolated into fleet-wide failure performance, and announced capacity plans were not treated as completed manufacturing output.

The geographic coverage follows the requested regions, groupings and featured countries rather than constituting a global deployment census. No customer interviews, factory visits or independent reliability tests were conducted. Public disclosure is uneven on shipments, yields, repair costs and installation locations. Named evidence gaps mean that findings remain unknown in this review—not that activity is absent. No market sizes, market shares or numeric forecasts were generated.

Our qualitative assessment is that CPO is becoming a credible option for demanding new AI fabrics, but the evidence does not support a universal replacement cycle for pluggable optics. A plausible near-term outcome is selective adoption alongside established optical architectures, with decisions governed by qualification, serviceability and supply readiness.

The milestones to watch are confirmed customer acceptance, reproducible package manufacturing, dependable repair workflows and interoperable operating configurations. Compute-adjacent optical I/O could extend integration further, but partnerships alone cannot establish that transition. Industry leaders should anchor commitments to product-level availability and operating evidence rather than to the scale of AI investment.