資料中心矽光電市場：規模、佔有率和成長預測（2026-2032 年）

預計到 2026 年，資料中心矽光電市場規模將達到 70 億美元，到 2032 年將達到 231 億美元。這意味著在預測期內，複合年成長率將達到 22.0%。

資料中心的光連結目前正經歷兩大變革。首先是速度從 400G 和 800G 連結到 1.6T 甚至 3.2T 的插入式光學模組的轉變。其次是架構上的變革，將光轉換功能更靠近交換器和運算晶片。矽光電在這兩大變革中都扮演著重要角色。在插入式模組中，矽光子積體電路 (PIC) 可以將多個調變器、檢測器和光路由功能整合到緊湊的收發器中。在近封裝、板載和共封裝實作中，同樣的整合原理使得光引擎能夠縮短高速電路徑，並在電損耗和重定時功耗過高之前將更多頻寬傳輸到光纖。因此，這項技術的應用範圍比共封裝光學模組更廣。思科在其 400G 到 1.6T 的插入式資料中心光學模組中採用了矽光電。相干公司正在為目前的 1.6T 模組和未來的 3.2T 模組開發矽光電PIC 實現方案。 GlobalFoundries 提供適用於各種外形規格的矽光電製程技術，包括插件式、線性驅動插件式、近封裝和共封裝。英特爾已出貨超過 800 萬顆矽光電PIC，其中包含超過 3,200 萬個用於資料中心收發器的片上雷射器，並正在將該平台擴展到 4Tbps 雙向光計算互連晶片。這個成熟的製造基礎至關重要，因為下一階段的部署將需要半導體級良率、測試自動化和封裝可重複性，而不是客製化光學組件。人工智慧正在改變光鏈路的經濟格局，因為頻寬需求成長速度遠超傳統伺服器流量。大規模GPU 叢集需要在交換器、機架和資料連接埠之間建立橫向擴展連結。更大規模的縱向擴展領域也越來越需要在加速器托架和機架之間建立光連接。多棟建築的人工智慧園區興起，也催生了對低成本「連貫光」光連接的需求，其傳輸距離已超出傳統資料中心的覆蓋範圍。矽光電可以透過不同的產品來滿足這些不同層面的需求。前面板模組支援橫向擴展。光引擎支援規模化擴展。連貫矽光電模組支援園區區間互聯。共封裝光元件支援交換器實現最高頻寬密度。因此，市場成長將得益於光端口數量的增加以及每個交換器或加速器系統中封裝的光電數量的增加。

市場促進因素

• 從 800G 到 1.6T 和 3.2T光學模組的轉變正在推動整合光電技術的廣泛應用。人工智慧橫向擴展網路速度的快速提升仍然是短期內出貨量的主要驅動力。思科推出了基於 Cisco Silicon Photonics 的 1.6T OSFP 平台，預計將於 2026 年上市。博通和 Marvell 也宣布了面向下一代模組的每通道 400G 和 1.6T 光矽技術。相干公司已展示了 1.6T 和 3.2T 的矽光電和磷化銦兩種實現方案，顯示市場將繼續保持多技術並存的局面。在通道數、整合密度和可製造性至關重要的領域，矽光電的競爭力日益增強。隨著速度的不斷提升，高速光通道的數量也在增加，將光功能整合到通用光子積體電路 (PIC) 上的價值也隨之提升。
• 由於電氣 I/O 的限制，光裝置正逐漸靠近交換器和加速器。對於大多數網路應用而言，前面板可拆卸模組仍然是可行的選擇。然而，隨著 SerDes 速度的提升，維護從交換器和加速器 ASIC 到光元件的電氣路徑變得越來越困難。重定時器和較長的基板走線會增加功耗和延遲。 NVIDIA 正透過其 Spectrum-X 乙太網路光子技術來解決這一限制，該技術將矽光電引擎直接放置在交換器 ASIC 旁邊，並計劃於 2026 年開始量產。 Lightmatter 則瞄準了不同的整合點，推出了近封裝板載光引擎 Passage L20。 Intel OCI 也在將光 I/O 整合到處理器封裝中。這些架構拓展了矽光電的潛在效益，使其不再侷限於取代傳統的收發器，而是在運算系統中創造了新的光相關元件。
• 人工智慧規模的擴大和多棟建築園區的建設正在為光纖通訊創造新的目標市場。人工智慧網路不再局限於單一資料連接埠內的乙太網路葉脊式連結。可擴展架構正在連接超越托盤和機架的大規模加速器域。超大規模園區橫跨多棟建築。 Marvell 正在開發用於人工智慧規模擴展的矽光電光引擎，以及用於 2 至 20 公里園區鏈路的連貫光解決方案。 Lightmatter 正在部署支援縱向擴展和橫向擴展的多Terabit光引擎。這些應用場景在以前只能使用銅線、電背板或更昂貴的連貫傳輸系統的地方增加了光鏈路。這提高了每兆瓦人工智慧運算所需的矽光電數量。
• 晶圓代工和製造生態係正接近量產階段。隨著製造模式從客製化光學組件轉向可重複使用的晶圓代工和封裝平台，矽光電將從中受益。 GlobalFoundries 表示，其成熟的矽光電製程支援插件式、近封裝式和共封裝式等多種外形規格。該公司預計，到 2026 年，其矽光電銷售額將成長一倍以上。該公司還宣布，可能獲得高達 3 億美元的 CHIPS 研發津貼，以加速下一代矽光電材料、晶圓和封裝的研發。 NVIDIA 正在利用台積電 (TSMC)、SPIL 和富士康 (Foxconn) 的資源，建構其 Spectrum-X Photonics 的矽光電量產供應鏈。這些投資將提高認證產能，並消除廣泛應用的關鍵障礙之一。

市場限制因素

• 矽光電不會取代所有競爭性的光學技術。磷化銦 (InP) 增強型調變雷射 (EML) 在高速插件模組方面仍然具有優勢，尤其是在直接雷射和調變器的性能優勢超過整合優勢的領域。垂直腔面發射雷射 (VCSEL) 基鏈路在短距離傳輸方面仍具有成本效益。矽光電也帶來了與雷射整合、光纖鍵合、熱控制和晶片良率 (KGD) 測試相關的製造挑戰。在共封裝光學元件和處理器鄰近光學元件中，由於光學系統故障可能影響比可更換的前面板模組更廣泛的系統，因此可維護性問題尤其突出。模組的平均售價通常會隨著每一代速度的模組實現量產而下降。因此，該預測預計產品數量和頻寬將強勁成長，但由於良率提高、競爭加劇和每位元傳輸成本降低，預計 2030 年後營收成長將逐漸放緩。

目錄

第1章：執行摘要

• 市場機會與主要發現
• 矽光電實施計劃
• 主要收入來源

第2章 市場概覽

• 資料中心矽光電的發展
• 可插拔、NPO、OBO 和 CPO 架構
• 矽光電與 InP EML 和 VCSEL 方法的比較
• 光輸入/輸出和處理器鄰近光電

第3章 市場規模及預測（2026-2032 年）

• 全球市場規模
• 年增率分析
• 每個光端口的矽光電含量
• 按插件和整合架構分類的收入

第4章 市場：依產品和整合形式分類

• 矽光電式收發器
• 光子積體電路和光引擎
• 近封裝光學元件
• 機載光學元件
• 共封裝光學引擎
• 光計算互連裝置
• 與晶圓代工廠和光子封裝相關的內容

第5章 市場：按角色互聯

• AI橫向擴展網路
• 人工智慧規模化網路
• 資料中心之間的互連和可擴展性
• 處理器和記憶體的光I/O

第6章 市場：依網路速度分類

• 400G 或更少
• 800G
• 1.6T
• 3.2噸或以上

第7章 市場：依客戶類型分類

• 超大規模雲端供應商
• Neocloud 和 GPU 雲端供應商
• 資料中心和網路OEM
• 主權人工智慧和高效能運算
• 企業資料中心

第8章 區域市場

• 北美洲
  • 美國
  • 加拿大
• 亞太地區
  • 中國
  • 台灣
  • 日本
  • 韓國
  • 東南亞
• 歐洲
• 中東和世界其他地區

第9章 市場動態

• 促進因素
  • 向 1.6T 和 3.2T光學模組遷移
  • 電氣 I/O 的功率和範圍限制
  • 人工智慧規模化和網路擴展
  • 晶圓代工與封裝生態系的成熟度
• 抑制因子
  • InP EML 和 VCSEL 架構之間的競爭
  • 雷射整合和光纖融合的複雜性
  • 整合光學的保守性
  • 每比特成本更低

第10章 競爭格局

• 網路和加速器平台供應商
• 光學元件和模組供應商
• 矽光電代工廠
• 光學 I/O 專家
• 包裝和製造生態系統

第11章：公司簡介

• NVIDIA
• Broadcom
• Marvell Technology
• Cisco
• Coherent
• Lumentum
• Intel
• GlobalFoundries
• TSMC
• Lightmatter
• Ayar Labs
• Tower Semiconductor
• Fabrinet
• ASE Technology
• Amkor Technology

第12章 近期趨勢

第13章附錄

The Data Center Silicon Photonics Market is estimated at USD 7.0 billion in 2026 and is projected to reach USD 23.1 billion by 2032, representing a CAGR of 22.0% during the forecast period.

Data-center optical connectivity is moving through two transitions simultaneously. The first is the speed transition from 400G and 800G links toward 1.6T and later 3.2T pluggable optics. The second is an architectural transition that moves the optical conversion closer to switch and compute silicon. Silicon photonics participates in both. In pluggable modules, a silicon PIC can integrate multiple modulators, detectors and optical routing functions into a compact transceiver. In near-package, on-board and co-packaged implementations, the same integration principle allows optical engines to shorten high-speed electrical paths and move more bandwidth onto fiber before electrical loss and retiming power become excessive. The technology is therefore broader than co-packaged optics. Cisco is using silicon photonics in 400G to 1.6T pluggable data-center optics. Coherent is developing silicon-photonics PIC implementations for both current 1.6T and future 3.2T modules. GlobalFoundries positions its silicon-photonics process technologies across pluggable, linear-drive pluggable, near-package and co-packaged form factors. Intel has already shipped more than eight million silicon-photonics PICs with more than 32 million on-chip lasers in data-center transceivers and is extending the platform into a 4 Tbps bidirectional optical compute interconnect chiplet. This installed manufacturing base matters because the next phase of adoption requires semiconductor-like yield, test automation and packaging repeatability rather than bespoke optical assembly. AI changes the economics of optical links because bandwidth demand is increasing faster than general server traffic. Large GPU clusters require scale-out links between switches, racks and data halls. Larger scale-up domains increasingly require optical connectivity between accelerator trays and racks. The rise of multi-building AI campuses also creates demand for lower-cost coherent-lite optical connections across distances beyond conventional data-center reach. Silicon photonics can address each of these layers with different products. Front-panel modules serve scale-out. Optical engines serve scale-up. Coherent silicon-photonics modules serve campus connectivity. Co-packaged optics serves the highest switch bandwidth density. The result is a market that grows through both higher optical port volumes and increasing photonic content per switch or accelerator system.

Market Drivers

• Transition from 800G to 1.6T and 3.2T optics increases integrated photonic content. The near-term volume driver remains the rapid speed upgrade in AI scale-out networks. Cisco launched a 1.6T OSFP platform in 2026 using Cisco Silicon Photonics. Broadcom and Marvell introduced 400G-per-lane and 1.6T optical silicon intended for the next generation of modules. Coherent is demonstrating both silicon-photonics and indium-phosphide implementations for 1.6T and 3.2T. This shows that the market will remain multi-technology. Silicon photonics is increasingly competitive where high lane count, integration density and manufacturability matter. Each speed transition raises the number of high-speed optical lanes and strengthens the value of integrating optical functions on a common PIC.
• Electrical I/O limits are pushing optics closer to switches and accelerators. Front-panel pluggable modules remain practical across a large portion of the network. Their electrical path from the switch or accelerator ASIC becomes harder to maintain as SerDes speeds rise. Retimers and long printed-circuit-board traces add power and latency. NVIDIA is addressing this constraint through Spectrum-X Ethernet Photonics. This places silicon-photonics engines beside the switch ASIC and entered production in 2026. Lightmatter is targeting a different integration point through Passage L20 near-package and on-board optical engines. Intel OCI similarly moves optical I/O to the processor package. These architectures expand silicon-photonics revenue beyond conventional transceiver replacement and create new optical content inside compute systems.
• AI scale-up and multi-building campuses add new optical addressable markets. AI networking is no longer limited to Ethernet leaf-spine links inside one data hall. Scale-up fabrics increasingly connect larger accelerator domains across trays and racks. Hyperscale campuses span multiple buildings. Marvell is developing silicon-photonics light engines for AI scale-up and coherent-lite solutions for 2 to 20 kilometre campus links. Lightmatter is positioning multi-terabit optical engines for both scale-up and scale-out. These use cases add optical links in places that were previously served by copper, electrical backplanes or more expensive coherent transport systems. This raises the silicon-photonics content per megawatt of AI compute.
• Foundry and manufacturing ecosystems are becoming production ready. Silicon photonics benefits when manufacturing moves from custom optical assembly toward reusable foundry and packaging platforms. GlobalFoundries states that its production-proven silicon-photonics processes support pluggable, near-package and co-packaged form factors. It expects its silicon-photonics revenue to more than double in 2026. The company also announced a potential USD 300 million CHIPS Research and Development award to accelerate next-generation silicon-photonics materials, wafers and packaging. NVIDIA is using TSMC, SPIL and Foxconn in a production silicon-photonics supply chain for Spectrum-X Photonics. These investments increase qualified capacity and reduce one of the principal barriers to broad adoption.

Market Restraints

• Silicon photonics does not displace every competing optical technology. Indium phosphide EMLs remain strong in high-speed pluggables, particularly where direct laser and modulator performance outweigh integration advantages. VCSEL-based links remain cost-effective at shorter reaches. Silicon photonics also introduces manufacturing challenges around laser integration, fiber attachment, thermal control and known-good-die testing. Co-packaged optics and processor-adjacent optics add serviceability concerns because optical failures can affect a larger system than a replaceable front-panel module. Module average selling prices typically decline after each speed generation reaches volume production. The forecast therefore assumes strong unit and bandwidth growth but progressively slower revenue growth after 2030 as yields improve, competition increases and cost per transmitted bit falls.

Technology and Segment Insights

• Silicon-photonics-based pluggable transceivers represent the largest revenue pool in 2026. 400G and 800G deployments are already at scale. 1.6T is beginning a major AI-driven ramp. These modules combine a silicon PIC with lasers, electronic drivers or DSPs, fiber interfaces and packaging in a field-replaceable form factor. Optical engines and PIC subsystems used in near-package, on-board and co-packaged architectures represent the fastest-growing category through 2032. Their starting base is smaller. Adoption expands rapidly as switch and accelerator electrical I/O approaches practical power and reach limits. Silicon-photonics foundry and specialized packaging revenue grows alongside both categories. It remains a smaller direct revenue pool than finished optical subsystems.
• Scale-out networking remains the largest application because Ethernet and InfiniBand fabrics connect very large numbers of servers, switches and accelerator nodes using optical links between racks and data halls. Scale-up networking is the fastest-growing role as accelerator domains extend beyond a single rack. Optical engines begin replacing copper for higher-bandwidth, longer-reach links. Data-center interconnect and scale-across applications form a separate high-value category that increasingly uses coherent silicon photonics for campus and metro distances. Processor optical I/O remains early-stage in 2026 but becomes increasingly relevant later in the forecast period as optical chiplets move from evaluation toward production systems.
• Silicon-photonics pluggables form the largest revenue pool with strong growth. Their primary demand is the 400G and 800G installed base and the 1.6T transition. Near-package and on-board optics optical engines are at early commercial stage with very high growth. Their primary demand is shorter electrical paths for switches and XPUs. Co-packaged optical engines are at initial production stage with very high growth. Their primary demand is the highest switch bandwidth density and lower electrical I/O power. Optical compute I/O is an emerging category accelerating after 2028. Its primary demand is processor-adjacent optical links and resource disaggregation. Scale-out networking is the largest use case with strong growth. Its primary demand is AI switch-to-switch and switch-to-network-interface connectivity. Scale-up networking has a smaller 2026 base but is the fastest-growing use case. Its primary demand is multi-rack accelerator domains and copper-reach limits.
• The market is segmented by product and integration form, interconnect role, link speed, customer type and geography. Product and integration forms include silicon-photonics pluggable transceivers, photonic integrated circuits and light engines, near-package optics, on-board optics, co-packaged optical engines, optical compute interconnect devices, and foundry and photonic packaging content. Interconnect roles include AI scale-out networking, AI scale-up networking, data center interconnect and scale-across, and processor and memory optical I/O. Link speeds include 400G and below, 800G, 1.6T, and 3.2T and higher. Customer types include hyperscale cloud providers, neocloud and GPU-cloud operators, data center and network OEMs, sovereign AI and HPC, and enterprise data centers.
• Silicon photonics expands from pluggable transceivers into near-package, on-board and co-packaged optical architectures. Cisco is introducing 1.6T pluggable optics using its silicon-photonics technology for AI networks. NVIDIA Spectrum-X Ethernet Photonics entered production during 2026 for Vera Rubin AI factories. GlobalFoundries expects silicon-photonics revenue to more than double during calendar year 2026. Coherent is demonstrating silicon-photonics implementations across 1.6T, 3.2T and future optical modules. Pluggable transceivers remain the largest 2026 revenue pool despite growing integrated-optics adoption. Optical engines and processor-adjacent photonics are the fastest-growing silicon-photonics revenue category. AI scale-out remains the largest networking use case while optical scale-up expands rapidly. North America leads design and early deployment. Integrated lasers, external laser sources and fiber attachment remain key architectural differentiation points.

Competitive and Strategic Outlook

• Competition spans four layers. Networking and accelerator-platform companies such as NVIDIA, Broadcom, Marvell and Cisco increasingly control the architecture and qualification path for silicon-photonics products. Optical-component and module companies such as Coherent and Lumentum compete across lasers, silicon-photonics PICs, transceivers and optical subsystems. Foundries and packaging companies such as GlobalFoundries and TSMC provide process technology and manufacturing scale. Specialist photonics companies such as Lightmatter and Ayar Labs target optical I/O architectures that move photonics closer to compute.
• The competitive boundary between these layers is becoming less distinct. Broadcom and Marvell combine high-speed electronic silicon with optical platforms. Cisco combines switching, silicon photonics and coherent optics through the Acacia organization. GlobalFoundries is moving from wafer processing into reference optical-module solutions. NVIDIA coordinates foundry, packaging, laser, fiber and system partners around a complete photonics switch platform. Differentiation therefore depends not only on modulator performance or optical bandwidth. It depends on packaging yield, laser strategy, fiber attachment, power per bit, field serviceability, interoperability and the ability to qualify an end-to-end system at hyperscale volumes.
• North America is the largest early demand and design center for data-center silicon photonics. NVIDIA, Broadcom, Marvell, Cisco, Intel, Coherent, Lumentum, GlobalFoundries, Lightmatter and Ayar Labs all operate major silicon-photonics, optical-networking or optical-I/O programs from the United States. The region also contains the largest concentration of hyperscale and AI-cloud buyers that can justify rapid adoption of new optical architectures. Networking power, GPU utilization and deployment speed directly affect the economics of AI infrastructure. Cisco is moving 1.6T silicon-photonics pluggables into AI scale-out networks. NVIDIA is commercializing co-packaged optics switches. Lightmatter is targeting optical engines for multi-rack scale-up.
• Manufacturing capacity is also becoming strategically important in the region. GlobalFoundries is expanding silicon-photonics and silicon-germanium capacity in the United States. In July 2026, it announced a letter of intent for a USD 300 million U.S. Department of Commerce award to advance silicon-photonics wafer technology, materials and packaging. The company reported seven new optical-networking design wins during the second quarter and expects silicon-photonics revenue to more than double in 2026. This foundry expansion complements domestic optical-module, laser and packaging investment. It reduces dependence on a single manufacturing geography for critical AI interconnect components.
• North American adoption is expected to remain heterogeneous. Pluggable optics continue to dominate high-volume switch-to-switch and switch-to-network-interface connectivity. Co-packaged optics is introduced first where switch bandwidth density and power justify deeper integration. Near-package and on-board engines can provide an intermediate architecture where serviceability is important. Optical compute I/O represents the longer-term opportunity as accelerators, memory and disaggregated resources require more bandwidth than electrical package interfaces can provide. This mix supports broad silicon-photonics growth without assuming an abrupt replacement of pluggable transceivers.
• Asia Pacific is critical to silicon-photonics manufacturing through foundry, outsourced semiconductor assembly and test, optical-module and server manufacturing capacity in Taiwan, China, Malaysia, Singapore, Japan and South Korea. TSMC and regional packaging partners are central to NVIDIA and other high-volume photonics programs. Chinese optical-module suppliers represent a large share of global datacom production. Europe contributes important photonics design, equipment and research capabilities. It remains a significant market for hyperscale, sovereign AI and high-performance computing infrastructure. Middle Eastern AI campuses are primarily demand centers for imported optical systems. Large greenfield deployments can adopt 1.6T, co-packaged optics and optical scale-up architectures rapidly because network designs are being built from the beginning around high-density accelerator systems.
• Recent developments confirm accelerating market activity. In September 2026, Lightmatter introduced the Passage L20 CPX bidirectional optical engine for AI scale-up networks under the Open CPX MSA. In July 2026, GlobalFoundries announced a letter of intent for a USD 300 million U.S. CHIPS Research and Development award focused on next-generation silicon photonics and packaging. In May 2026, NVIDIA announced that Spectrum-X Ethernet Photonics had entered full production with the Vera Rubin platform. In May 2026, GlobalFoundries introduced its SCALE silicon-photonics co-packaged optical module platform for AI scale-up architectures. In March 2026, Coherent demonstrated 1.6T silicon-photonics transceivers and 400G-per-lane silicon-photonics links for future 3.2T pluggables. In February 2026, Cisco announced 1.6T OSFP and 800G linear pluggable optics using Cisco Silicon Photonics for AI data-center networking.

Conclusion

• The data center silicon photonics market is growing rapidly as optical connectivity moves closer to switch and compute silicon. Growth is driven by the 1.6T and 3.2T optical transition, electrical I/O power and reach limits, expansion of AI scale-up networking and foundry and packaging ecosystem maturity. Silicon-photonics pluggables form the largest revenue pool. Optical engines and processor-adjacent photonics are the fastest-growing category. Scale-out networking remains the largest use case. Scale-up networking is the fastest-growing use case. North America leads design and early deployment. Asia Pacific remains critical to manufacturing. Competition from indium phosphide and VCSEL architectures, laser integration complexity and serviceability remain the key restraints. The market is projected to grow from USD 7.0 billion in 2026 to USD 23.1 billion by 2032 at a CAGR of 22.0%.

Key Benefits of this Report

• Insightful Analysis: Detailed market insights across regions, customer segments, policies, socio-economic factors, consumer preferences, and industry verticals.
• Competitive Landscape: Understand strategic moves by key players to identify optimal market entry approaches.
• Market Drivers and Future Trends: Assess major growth forces and emerging developments shaping the market.
• Actionable Recommendations: Support strategic decisions to unlock new revenue streams.
• Caters to a Wide Audience: Suitable for startups, research institutions, consultants, SMEs, and large enterprises.

What Businesses Use Our Reports For

• Industry and market insights, opportunity assessment, product demand forecasting, market entry strategy, geographical expansion, capital investment decisions, regulatory analysis, new product development, and competitive intelligence.

Report Coverage

• Historical data from 2021 to 2024, Base year 2025, and Forecast years from 2026 to 2032
• Growth opportunities, challenges, supply chain outlook, regulatory framework, and trend analysis
• Competitive positioning, strategies, and market share evaluation, and trade analysis
• Revenue growth and forecast assessment across segments and regions
• Company profiling including strategies, products, financials, and key developments

TABLE OF CONTENTS

1. Executive Summary

• 1.1. Market Opportunity and Key Findings
• 1.2. Silicon Photonics Adoption Timeline
• 1.3. Principal Revenue Pools

2. Market Overview

• 2.1. Evolution of Silicon Photonics in Data Centers
• 2.2. Pluggable, NPO, OBO and CPO Architectures
• 2.3. Silicon Photonics versus InP EML and VCSEL Approaches
• 2.4. Optical I/O and Processor-Adjacent Photonics

3. Market Size and Forecast, 2026-2032

• 3.1. Global Market Revenue
• 3.2. Annual Growth Analysis
• 3.3. Silicon Photonics Content per Optical Port
• 3.4. Revenue by Pluggable and Integrated Architectures

4. Market by Product and Integration Form

• 4.1. Silicon-Photonics Pluggable Transceivers
• 4.2. Photonic Integrated Circuits and Light Engines
• 4.3. Near-Package Optics
• 4.4. On-Board Optics
• 4.5. Co-Packaged Optical Engines
• 4.6. Optical Compute Interconnect Devices
• 4.7. Foundry and Photonic Packaging Content

5. Market by Interconnect Role

• 5.1. AI Scale-Out Networking
• 5.2. AI Scale-Up Networking
• 5.3. Data Center Interconnect and Scale-Across
• 5.4. Processor and Memory Optical I/O

6. Market by Link Speed

• 6.1. 400G and Below
• 6.2. 800G
• 6.3. 1.6T
• 6.4. 3.2T and Higher

7. Market by Customer Type

• 7.1. Hyperscale Cloud Providers
• 7.2. Neocloud and GPU-Cloud Operators
• 7.3. Data Center and Network OEMs
• 7.4. Sovereign AI and HPC
• 7.5. Enterprise Data Centers

8. Regional Market

• 8.1. North America
  • 8.1.1. United States
  • 8.1.2. Canada
• 8.2. Asia Pacific
  • 8.2.1. China
  • 8.2.2. Taiwan
  • 8.2.3. Japan
  • 8.2.4. South Korea
  • 8.2.5. Southeast Asia
• 8.3. Europe
• 8.4. Middle East and Rest of World

9. Market Dynamics

• 9.1. Drivers
  • 9.1.1. 1.6T and 3.2T Optical Transition
  • 9.1.2. Electrical I/O Power and Reach Limits
  • 9.1.3. Expansion of AI Scale-Up Networking
  • 9.1.4. Foundry and Packaging Ecosystem Maturity
• 9.2. Restraints
  • 9.2.1. Competition from InP EML and VCSEL Architectures
  • 9.2.2. Laser Integration and Fiber-Attach Complexity
  • 9.2.3. Serviceability of Integrated Optics
  • 9.2.4. Declining Cost per Bit

10. Competitive Landscape

• 10.1. Networking and Accelerator Platform Vendors
• 10.2. Optical Component and Module Suppliers
• 10.3. Silicon Photonics Foundries
• 10.4. Optical I/O Specialists
• 10.5. Packaging and Manufacturing Ecosystem

11. Company Profiles

• 11.1. NVIDIA
• 11.2. Broadcom
• 11.3. Marvell Technology
• 11.4. Cisco
• 11.5. Coherent
• 11.6. Lumentum
• 11.7. Intel
• 11.8. GlobalFoundries
• 11.9. TSMC
• 11.10. Lightmatter
• 11.11. Ayar Labs
• 11.12. Tower Semiconductor
• 11.13. Fabrinet
• 11.14. ASE Technology
• 11.15. Amkor Technology

12. Recent Developments

13. Appendix