Silicon photonics Market Outlook 2026–2034: Growth Drivers, Key Companies, Strategic Developments & Adoption Forecast — A Meticulous Next™ Foresight Brief

This Meticulous Next™ brief examines how silicon photonics and co-packaged optics, known as CPO, change the networks inside AI data centers over the next eight years. The change moves the optical engine from a pluggable module at the edge of a switch to the same package as the switch chip, and it responds to the bandwidth and power limits of clusters that link tens of thousands of accelerators.

The case begins with power. Broadcom states that its third-generation CPO switch cuts optical interconnect power by 70%, more than 3.5 times lower than traditional pluggable solutions. In a cluster where networking can take a meaningful share of power, a reduction of that size frees electricity for compute. That makes the case for CPO a case about the cost of running AI, and not only about speed.

The first proof is product. On 8 October 2025 Broadcom announced Tomahawk 6 Davisson, a 102.4 terabit per second Ethernet switch with co-packaged optics built on TSMC COUPE optical engines, supporting scale-up clusters of 512 accelerators and two-tier networks of more than 100,000. Nvidia described its Spectrum-X Ethernet Photonics switches as in production on 31 May 2026, with CoreWeave, Lambda and Oracle Cloud Infrastructure named as early partners. Two leading suppliers now ship, which turns CPO from a demonstration into a purchasing choice.

The second proof is the foundry route. TSMC set out in April 2026 a first CPO configuration with COUPE on substrate, planned to begin production during 2026, and Samsung presented a 300-millimetre silicon photonics platform at OFC 2026. When the largest foundries commit, photonic chips gain the scale and yield discipline of mainstream manufacturing.

The third proof is the constraint. Lasers, fibre, connectors and test are emerging as bottlenecks. Lumentum chief executive described a gap above 30% between laser shipments and demand in May 2026, and Coherent began expanding an indium phosphide fab in Sherman, Texas in June 2026. A technology that depends on specialist materials will scale at the pace of its slowest input, which is why suppliers of lasers and test equipment gain pricing power.

The brief maps the technology, its indicative market size and forecast, the factors behind its growth, the developments of the last 24 months, the key players and the adoption trajectory to 2034.

It is a focused 110-page decision brief for data center operators, switch and accelerator makers, optical vendors, foundries, laser and materials suppliers, investors and policy makers. It presents an indicative trajectory rather than a segmented market model. Its purpose is to identify which links adopt CPO first, where supply limits persist, and who captures value as optics move into the package.

Silicon photonics builds optical components such as modulators, waveguides and detectors on silicon wafers using semiconductor processes. Light is generated by a laser, often made of indium phosphide, and either bonded to the chip or supplied from an external source. The gain is that optical parts can be made at the volume and cost of electronic chips.

Co-packaged optics places the optical engine on the same package as the switch or accelerator. Electrical signals travel a few millimetres instead of across a board to a pluggable module, which reduces power and loss. Broadcom third-generation design integrates sixteen 6.4 terabit optical engines, uses field-replaceable laser modules and follows IEEE 802.3, with a fourth generation planned at 400 gigabits per channel.

The packaging layer is the second constraint. Optical engines must be combined with electronic dies on advanced packages, using processes such as TSMC COUPE, and must keep alignment and thermal stability. This ties CPO to the capacity of advanced packaging described in other briefs in this series.

Lasers, fibre and test are the third constraint. Indium phosphide substrates and laser capacity are tight, fibre attach and connectors need new designs, and testing a photonic die can take over 100 seconds. Known-good die testing has no clear owner yet, which is an opportunity for equipment suppliers.

The limits are serviceability, supply and standards. A failed pluggable module can be swapped in minutes, whereas a failed engine on a package is harder to replace, which is why laser modules are made replaceable. The decade turns on whether yield, laser supply and test scale together and whether operators accept new maintenance models.

The silicon photonics market, covering photonic integrated circuits, optical transceivers and engines, co-packaged optics, external laser sources, and related packaging, fibre and test, and excluding electrical interconnects, traditional discrete optics and fibre cable, is valued at USD 4–6 billion in 2026 and is projected to reach USD 25–38 billion by 2034, at a CAGR of 23%.

The 2026 base rests on silicon photonics transceivers for data centers, first CPO switches and the foundry and test capacity that supports them. From that base, growth follows the size of AI clusters. Each step in cluster size multiplies the number of optical links, and each generation of switch raises the share carried by CPO.

The forecast is sensitive to AI infrastructure spending, to laser supply and to the speed at which operators accept CPO. Regionally, the United States leads on design and cloud demand, Taiwan on foundry and packaging, and China, Japan and Europe on transceivers, materials and components.

The base case assumes AI infrastructure investment remains high, laser supply expands steadily and operators accept CPO in the largest clusters, giving the inflection around 2029. An accelerated case adds faster laser capacity and early optical I/O for accelerators, bringing the inflection to 2028 and lifting the 2034 value to the top of the range. A delayed case assumes slower AI spending, persistent laser shortages and reliability concerns, holding growth at the bottom of the range.

Growth drivers

• Bandwidth demand from AI clusters that link tens of thousands of accelerators.
• Power savings, with claims of a 70% reduction in optical interconnect power.
• Foundry commitment to photonics platforms.
• Standards such as IEEE 802.3 support for CPO links.

Enablers

• Switch products from Broadcom and Nvidia that ship in volume.
• Advanced packaging such as TSMC COUPE.
• Replaceable external laser modules.
• Public and private investment in indium phosphide capacity.

Restraints and barriers

• Shortage of lasers and indium phosphide substrates.
• Test and known-good die gaps.
• Serviceability concerns compared with pluggable modules.
• Dependence on advanced packaging capacity.

Five converging forces will determine how fast silicon photonics and CPO scale.

(1) AI cluster bandwidth demand. (2) Power cost. (3) Foundry and packaging capacity. (4) Laser and materials supply. (5) Operator acceptance and standards. The brief assesses each force for direction, speed and confidence.

The key players operating in silicon photonics and co-packaged optics include Broadcom Inc., NVIDIA Corporation, Taiwan Semiconductor Manufacturing Company Limited, Intel Corporation, Samsung Electronics Co., Ltd., GlobalFoundries Inc., Tower Semiconductor Ltd., Marvell Technology, Inc., Cisco Systems, Inc., Arista Networks, Inc., Juniper Networks, Inc., Ayar Labs, Inc., Lightmatter, Inc., Celestial AI Inc., Lumentum Holdings Inc., Coherent Corp., Corning Incorporated, Sumitomo Electric Industries, Ltd., AXT, Inc., Fabrinet, Innolight Technology Corporation, Eoptolink Technology Inc., Ltd., Applied Optoelectronics, Inc., Ciena Corporation, Nokia Corporation, Celestica Inc., Hewlett Packard Enterprise Company, Advanced Semiconductor Engineering, Inc., Amkor Technology, Inc., SENKO Advanced Components, Inc., Teradyne, Inc., Advantest Corporation, Microsoft Corporation, Alphabet Inc., Amazon.com, Inc., Meta Platforms, Inc., Oracle Corporation, CoreWeave, Inc., the Open Compute Project Foundation and the IEEE.

The brief profiles representative players in each archetype and assesses which are positioned to lead as optics move into the package.

The competitive landscape is forming around six archetypes. Switch and accelerator makers. Foundries and packagers. Photonic engine specialists. Laser, fibre and materials suppliers. Transceiver and network equipment makers. Data center operators and standards bodies. Competitive intensity is high in 2026 and is expected to rise as foundries and start-ups reach production through 2029.

Where value migrates.

In 2026 value sits in switch silicon, foundry platforms and the laser and materials that limit supply. By 2029 it moves to packaging and test, external laser sources and fibre attach as volumes rise. By 2034 it settles in owners of integrated platforms, optical I/O for accelerators, and suppliers of lasers and test with durable capacity. Pluggable transceiver margins fall as CPO takes the largest links.

The signals converge on shipment and supply. Two leading suppliers have CPO switches in production, foundries have committed to photonics platforms, and laser capacity is being added. Each signal moves CPO from a demonstration to a standard option in the largest clusters.

Research and patent activity reflects where the chain is weakest: laser integration, fibre attach, known-good die test, thermal control and optical I/O for accelerators. The brief therefore tracks four indicators: share of switch ports using CPO, laser output against demand, photonic die test time and the number of accelerators with optical I/O.

Regionally, the United States leads on design and cloud demand, Taiwan on foundry and packaging, and China, Japan and Europe on transceivers, materials and components. Scale-out switching, scale-up accelerator links, storage networks and telecom edge are the first demand pools in that order.