Last Updated on 2026 年 4 月 1 日 by 総合編集組
1.6T Optical Transceiver Mass Production Turning Point: How 400G per Lane Technology, Silicon Photonics, and CPO Are Reshaping AI Data Center Interconnects
Introduction The optical communications industry in 2026 stands at a critical inflection point driven by explosive demand for artificial intelligence infrastructure. As generative AI models require ever-higher bandwidth inside data centers, 1.6T optical transceivers have officially moved from sampling to large-scale volume production ramp. This shift focuses not only on speed but also on power efficiency, packaging density, and system reliability, making optical interconnects a core enabler for scalable AI factories.
Market Scale and the Rise of the 1.6T Supercycle Global high-end optical transceiver market is projected to reach 16.22 billion USD in 2026, fueled by massive capital expenditures from hyperscalers such as Amazon, Google, Meta, and Microsoft. Their combined capex approached 350 billion USD in 2025 and continues to expand in 2026, with a significant portion allocated to optical networking to support growing XPU interconnect needs.
Shipment forecasts show dramatic growth: 1.6T module shipments are expected to surge from 2.5 million units in 2025 to over 20 million units in 2026, continuing upward to 35 million in 2027 and 50 million in 2028. By 2030, the market could exceed 70 million units with a total value near 35 billion USD, marking the transition toward all-optical networks where copper cables inside racks are fully replaced.
This “supercycle” emerges because traditional 400G and 800G networks can no longer meet the low-latency requirements of trillion-parameter large models, forcing operators to upgrade before hitting the power wall and IO wall.
Core Technology Breakthroughs: 400G per Lane and Silicon Photonics Integration At the heart of 1.6T technology is the move to 224G PAM4 per lane, with leading vendors already demonstrating 400G per lane prototypes at OFC 2026. Higher lane rates directly influence module packaging and power consumption.
Broadcom leads with its Taurus series (BCM83640), the industry’s first 400G/lane optical DSP, paired with EML lasers and photodiodes to enable ultra-low-power 1.6T transceivers and lay groundwork for 3.2T solutions. Signal integrity challenges at high frequencies demand advanced adaptive equalization algorithms in DSP.
Silicon photonics has become mainstream, expected to capture 60-70% market share in 1.6T transceivers in 2026 due to compatibility with mature CMOS processes and cost advantages. TSMC and Samsung are investing in 300mm silicon photonics platforms for highly integrated packaging. Thin-film lithium niobate (TFLN) is also gaining attention for its low drive voltage and linearity at extreme frequencies, though integration challenges with silicon photonics remain.
Key Highlights from OFC 2026 The OFC 2026 conference in Los Angeles served as a major industry benchmark. Coherent, a founding member of the XPO MSA, showcased innovations for 12.8T liquid-cooled optical modules to address thermal management in future AI clusters, along with 400G/lane 3.2T transceivers emphasizing vertical integration with InP CW lasers and VCSEL technology.
Lumentum reported strong momentum in cloud transceiver business, with 1.6T modules scheduled for summer 2026 shipments and internal laser chip integration to improve margins. Its Optical Circuit Switching (OCS) business has backlog exceeding 400 million USD, projecting quarterly revenue of 100 million USD by end of 2026, driven by the R300 series 300×300 high-radix systems.
Nvidia is deeply influencing the supply chain, announcing a 2 billion USD investment in Lumentum for capacity and R&D, while supporting Coherent’s upgrade of InP production lines from 3-inch to 6-inch wafers, quadrupling output per wafer. This direct funding model highlights optical components as strategic assets for AI compute delivery.
Co-Packaged Optics (CPO) and Next-Generation Architectures As rack power density approaches 600kW, traditional pluggable modules face severe thermal and power challenges, pushing the industry toward closer-to-chip packaging.
CPO integrates optical engines with switch ASICs on the same substrate, reducing electrical signal distance and cutting energy per bit from 15pJ to about 5pJ. In 1.6T networks, shifting from pluggable to CPO can lower link power from 30W to 9W.
Comparison of three major interconnect architectures:
- Pluggable: Mature ecosystem, easy replacement, but high power and signal loss at high frequencies. Dominant in mainstream market in 2026.
- Linear Drive (LPO/LRO): Removes DSP for lower latency and power; suitable for intra-rack applications, though limited reach and strict host requirements. Pioneered by Applied Optoelectronics with Microsoft adoption.
- CPO: Lowest power and highest density, but faces maintenance, testing, and standardization challenges. Currently in early pilots.
LPO serves as a practical transitional solution while CPO matures.
Latest Market Movements as of March 31, 2026 Despite solid technical fundamentals, capital markets showed volatility with risk-off behavior. Applied Optoelectronics secured a 53 million USD order from a North American hyperscaler for 800G and early 1.6T deployments, yet its stock dropped 14.2% that day, reflecting concerns over potential digestion periods in AI spending and profit-taking after prior rallies.
Lumentum, despite a 9% weekly decline, was added to the S&P 500, solidifying its position as a core AI optics supplier. Taiwanese vendor FICG announced stable mass production of 1.6T modules with exceptional first-pass yield exceeding 99.997% on 01005 miniature components, critical for signal stability in high-frequency environments.
Community discussions on Reddit and Hacker News focus on reliability: CPO’s non-repairable nature raises concerns about whole-switch failures, driving interest in External Laser Source (ELS) designs. LPO’s interoperability and host-side signal quality requirements also draw cautious views. Investors emphasize that the real transition remains 800G to 1.6T, with every InP laser chip in high demand.
Emerging Transmission Media and Regional Supply Chain Dynamics Hollow-Core Fiber (HCF) offers ~30% faster light speed, ultra-low latency, and reduced nonlinearity, with Microsoft and Google testing it in key links for large language model training efficiency.
Micro-LED based active optical cables (AOC) demonstrate superior energy efficiency over VCSEL for ultra-short reaches, combining copper-like reliability with better bandwidth and distance.
Regionally, North America leads in hyperscale data centers and R&D (Nvidia, Broadcom, Cisco). Asia-Pacific shows strongest manufacturing resilience, with Thailand becoming a major production base for Lumentum’s 800G/1.6T modules to mitigate geopolitical risks. Europe focuses on open disaggregated optical networks.
2026 Q2 Outlook and Challenges Key hurdles include laser chip supply gaps (especially EML), integration of optics with liquid cooling, and increasing export compliance risks for AI-related hardware. Performance targets for the next 18-24 months include pushing lane rates to 448 Gb/s PAM4, power below 1 pJ/bit, switch capacity to 204.8T, and packaging toward KPO/CPO sockets.
Conclusion Optical connectivity has evolved from a simple networking pipe into an active asset in AI compute clusters. Innovations from Broadcom’s DSP, Nvidia’s strategic investments, and the 1.6T production ramp are collectively breaking through the IO wall. While short-term market fluctuations exist due to valuations, the long-term direction toward optics moving closer to—or inside—chip packaging is irreversible.
Signal margin calculations often use the Shannon capacity formula C = B · log₂(1 + SNR), while advanced FEC is essential at 224G/lane to manage bit error rates.
Disclaimer: This summary is for informational purposes only based on public data as of March 31, 2026. It is not investment advice. Readers should conduct their own research and consult professionals.
