The competition for next-generation optical modules won't be a winner-takes-all scenario?

The explosive expansion of AI computing power is reshaping the landscape of the optical interconnect industry, but the outcome of this technological race will not see dominance by any one single route.

The latest industry report from Zheshang Securities notes that currently, the optical interconnect industry does not exhibit a pattern where a single technology comprehensively replaces other routes. Instead, it features the distinct characteristic of "application scenarios determining technology selection, multi-route division of labor and collaboration, and long-term coexistence."

As data center electrical interconnects face the triple bottleneck of bandwidth, latency, and power consumption, diverse technological routes such as silicon photonics, LPO, LRO, NPO, CPO, TFLN, etc. are evolving concurrently.

Betting on a single technology route in the optical module sector carries significant risk, while industry value is concentrating in upstream segments like optical chips, advanced packaging, and specialty optoelectronic materials—high-barrier areas. These core components will be key variables determining the upper limits of different technological routes.

Triple bottleneck spurs multiple parallel routes

Traditional electrical interconnect solutions are facing systemic failure. As single-node computing power breaks through 10^18 operations per second, copper-based electrical interconnects encounter the threefold challenge of the "bandwidth wall," "latency wall," and "power wall": single-channel rates struggle to exceed 400Gbps, transmission latency reaches several microseconds, and single rack interconnect power consumption exceeds 40% of the total.

Thus, optical interconnect technology has become the inevitable path forward, and the trend of "optical in, electrical out" is irreversible. But the issue is that optical interconnect itself is not monolithic. According to China Mobile's "White Paper on Optical Interconnect Technology for Large-Scale Intelligent Computing Clusters (2025)", optical interconnect technology can be divided into equipment-level and chip-level categories; the former is dominated by pluggable optical modules, while the latter includes NPO, CPO, and other near-packaging and co-packaging solutions.

Different technological routes have their own focuses in aspects such as power control, transmission latency, port bandwidth density, and device maintainability, which is the fundamental reason for the formation of multiple parallel routes. Zheshang Securities analysts Deng Hefang and Zhou Yixuan pointed out in the report that to clarify this pattern, it is necessary to comprehensively analyze the differentiated demands of various scenarios in terms of device maintainability, hardware standardization, and ecosystem maturity.

Silicon Photonics: A platform foundation with rising penetration

Silicon photonics is not a specific product, but a platform-level foundational technology for the entire optical interconnect field. Its core advantage lies in its high compatibility with CMOS processes, enabling ultra-large-scale mass production through mature foundries like TSMC, Intel, GlobalFoundries, and featuring ultra-high integration and powerful optoelectronic integration capabilities.

Market data validates this judgment. According to the LightCounting May 2026 report, 2026 will be a milestone year when the sales of transceivers using silicon photonic modulators surpass $4 billion, exceeding 50% of the total market. Yole Group predicts that the silicon photonics market will grow from $278 million in 2024 to approximately $2.7 billion by 2030, with a compound annual growth rate of 46%.

Looking at longer cycles, the optical chip market is projected to grow from $4 billion in 2025 to about $15 billion in 2031, with the silicon photonic chip share rising from the current one-third to 42%, corresponding to about $6.3 billion. Notably, the penetration path of silicon photonics will continue to extend with the evolution of optical interconnect architectures—expanding from the current scale-out horizontal network, gradually toward scale-up vertical expansion and even scale-in internal packaging networks.

Three-way differentiation within pluggable camp

Within the scope of pluggable optical modules, the industry is differentiating technologies by adjusting DSP configurations, forming three parallel routes: FRO (full DSP), LRO (semi-retiming), and LPO (fully linear).

LPO was launched in 2022 by Macom in collaboration with Nvidia. Its core logic is to completely remove the DSP chip, using a pure analog linear direct-drive architecture to significantly reduce power consumption and latency. According to Macom’s data, 800G multimode optical modules can reduce power consumption from over 13W to below 4W, with overall costs dropping by about 8%. However, LPO’s limitations are obvious: weak noise resistance, applicable scenarios limited to short-distance interconnects within 500 meters, lacking a unified interoperability standard, and requiring high SerDes performance on the system side.

LRO is a more pragmatic compromise. It retains a DSP only at the transmit side to ensure signal quality meets IEEE 802.3 standards, while the receive side uses a linear analog architecture to reduce power consumption. The IEEE Electronic Packaging Society’s March 2026 technical report indicates that when single-channel rates rise to 200G/lane and total module rate reaches 1.6Tbps, full DSP solution power consumption is expected to exceed 30W, while LRO can keep it below 20W—this threshold allows continued use of air cooling instead of liquid cooling, greatly reducing deployment complexity. The report also reveals that nearly all companies showcasing 1.6T LPO solutions at OFC 2025 are also presenting LRO solutions, and the industry generally believes LRO is more feasible than LPO in the 1.6T era.

NPO: The mainstream choice for current large-scale adoption

NPO (Near Packaging Optics) is positioned as a pragmatic transition solution between traditional pluggable and CPO. The core design is to mount the optical engine near the ASIC chip on the switch mainboard, shortening the electrical signal path to centimeters, greatly reducing insertion loss while maintaining replaceability of the optical engine.

NPO’s competitiveness lies in balancing performance and industry realities. Alibaba and Tencent tech experts believe that although CPO is optimal in terms of performance, the lack of an open ecosystem is a major concern; in contrast, NPO can rely on the mature pluggable optical module ecosystem while significantly improving in terms of bandwidth density and power consumption. Chen Qin, Alibaba Cloud Optical Network Architect, points out that at ≤224G/L rates, NPO has ample performance reserves and can fully reuse existing industry chains, making it easier for large-scale adoption. NPO is currently the main technical path chosen by domestic GPU chip manufacturers.

Market scale data supports this view. According to DataIntelo, the global near packaging optics market is estimated at $3.8 billion in 2025, with a compound annual growth rate of 19.3% expected from 2026 to 2034, reaching $18.6 billion by 2034. North America leads with a 36.2% market share, while the Asia-Pacific region is expected to chase with the fastest regional compound growth rate of 21.4%.

CPO: The ultimate direction, but commercialization challenges cannot be underestimated

CPO (Co-Packaged Optics) is recognized by the industry as the “ultimate solution.” Through 2.5D/3D advanced packaging technology, the optical engine and switch ASIC are integrated on the same substrate, compressing the electrical signal transmission path from over 100 mm in traditional solutions to millimeter scale, reducing power consumption by 30% to 50%, and achieving nanosecond-level ultra-low latency and over 3.2T+ bandwidth density per channel.

Nvidia and Broadcom are the most aggressive promoters of CPO. Nvidia released Quantum-X and Spectrum-X silicon photonic co-packaged chips at the 2025 GTC conference, with plans to deliver InfiniBand CPO systems in the first half of 2026; Broadcom delivered the industry’s first 51.2Tbps CPO Ethernet switch, Bailly, in March 2024. Its CPO production line is expected to enter critical mass production stage in the second half of 2026, and monthly output in the first quarter of 2027 may leap to ten thousand units. Market forecasts from LightCounting predict that the CPO market may reach $10 billion in 2030, and Coherent further raised this at OFC to $15 billion.

However, CPO’s commercialization challenges are not to be overlooked. On the technical side, CPO involves deep integration of chip design, photonic integration, advanced packaging, thermal management, and other fields, and the entire industry chain has yet to form a standardized system. The cost of a single optical engine can be as high as $35,000 to $40,000. On the maintenance side, CPO’s “non-pluggable” architecture permanently binds the optical engine and expensive ASIC—once a failure occurs, the entire composite module must be replaced, completely overturning the existing data center maintenance ecosystem. Furthermore, lack of interoperability consensus between Nvidia’s COUPE solution and Broadcom’s FOWLP solution, as well as the absence of industry standards, delays adoption.

TFLN: The new variable in high-end niche tracks

Thin-film lithium niobate (TFLN) as a new-generation optoelectronic material technology is opening up an independent track in high-end/high-speed optical modules. Lithium niobate crystals are referred to in the industry as “optical silicon”; their naturally low half-wave voltage enables modulators to be directly driven by DSP’s native low-swing electrical signals, without the need for external high-power driver amplification circuits.

Commercial breakthroughs have already occurred. TFLN-based 1.6T-DR8 optical transceivers operate at only 20 watts, 20% lower than comparable traditional solutions. Using a single continuous wave laser drive scheme greatly simplifies the optical path structure and maintenance complexity. HyperLight management points out that TFLN is the core supporting technology for future 400Gbps single-channel optical communication systems, and in the current mainstream 200Gbps technology generation, it demonstrates strong energy-saving capability.

The Zheshang Securities report judges that TFLN does not appear as a disruptive solution replacing existing mainstream technology, but as a key complementary technology filling gaps in high-performance electro-optical modulation. In the future, the industry will feature multiple technologies—silicon photonics, indium phosphide, TFLN—running in parallel and selected as needed, with TFLN maintaining a strong presence in high-end/high-speed optical modules and RF photonics device niches. TFLN has entered small-scale commercial deployment, and as manufacturing processes improve and mass production yield rates increase, it will gradually penetrate from high-end to general-purpose scenarios.

Risk Warning and DisclaimerThe market has risks, investments must be made with caution. This article does not constitute personal investment advice and does not take into account specific investment goals, financial situations, or needs of any individual user. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular circumstances. Investment is at your own risk.