1. Introduction: The Rise of AI Elevates Optical Modules to Strategic Importance
With the rapid rise of AI technologies, data has become a new production factor. The high-speed, low-latency, and energy-efficient flow of this data requires a robust communication infrastructure. In this transformation, optical transceivers—key components that convert electrical signals to optical signals and vice versa—are moving from behind the scenes to center stage, becoming the vital arteries supporting data centers, cloud computing, and AI training and inference.
2. AI Compute Boom Drives Upgrades in Optical Module Technology and Market
1) Exponential Bandwidth Demand
Large models like GPT4 now exceed one trillion parameters, with AI compute demand doubling every 3–4 months—far outpacing Moore’s Law. As a result, data center interconnect speeds are leaping from 100G to 800G, 1.6T, and beyond. For example, an NVIDIA DGX H100 server cluster requires at least 4.8 Tbps bandwidth per rack, demanding over 500 optical modules. The next-generation GB300 platform will further drive a wave of demand for new 1.6T modules.
2) Dual Challenges: Ultra-Low Latency and High Energy Efficiency
AI training and inference workloads are highly sensitive to latency (less than 100 nanoseconds), while traditional pluggable modules consume more than 30% of a data center’s total power. CPO (Co-Packaged Optics) can cut power consumption by about 40% and latency by around 50%, making it a key development direction for future data centers. By 2027, the CPO market is projected to exceed $5 billion.
3) Photonic Integration and Silicon Photonics Accelerate Commercialization
Silicon photonics integrates lasers, modulators, and detectors onto silicon chips using CMOS processes, significantly reducing cost and size. Intel and Cisco have released 400G silicon photonic modules, and domestic Chinese companies are ramping up production of 800G and 1.6T products. By 2030, the market share of silicon photonic modules is expected to rise from 20% in 2023 to over 60%.
3. Market Boom: Surging Shipments, Fierce Competition
1) Exponential Shipment Growth
TrendForce reports global shipments of 400G+ optical modules reached 6.4 million units in 2023, are expected to rise to 20.4 million in 2024, and exceed 31.9 million in 2025, with a 56.5% CAGR. This surge is fueled by cost reductions in AI models (e.g., DeepSeek), expanding cloud and edge AI applications, and accelerating construction of both micro and hyperscale data centers worldwide.
2) Evolving Global Market and Regional Structure
According to ResearchAndMarkets, the global optical module market will reach $11.5 billion in 2025, growing to $47.6 billion by 2035, with a 13.8% CAGR. Mordor Intelligence reports rapid growth in China and North America, with the Asia-Pacific region leading global momentum. In 2024, North America accounts for 62% of the global high-speed optical module market, but Asia-Pacific is catching up fast. On average, 75% of revenue for Chinese optical module makers comes from overseas: 86.8% for InnoLight, 78% for Eoptolink. Southeast Asia is a key manufacturing hub, with monthly output capacity exceeding 1.5 million units, including:
InnoLight: 500K
Eoptolink: 300K
HGTech: 200K
3) High-Speed Modules Drive Growth
The >400G segment is rapidly expanding. 1.6T module shipments are projected to grow from 2.7 million in 2024 to 4.2 million in 2025. 800G modules are also seeing explosive growth. Major cloud service providers (CSPs)—Meta, Microsoft, Amazon, and Google—are significantly increasing capital expenditures. The share of AI compute is rising, with costs of optical modules and PCBs increasing accordingly.
4) AI Investment Outpaces Optical Module Growth
According to CoWoS, AI chip demand will grow at 58% CAGR from 2024–2026, far outpacing cloud providers’ capex growth (~25%). The penetration of ASIC chips further drives optical module demand. By 2025, optical modules are expected to account for 18% of AI infrastructure costs, up from 12% in 2023. For example, Meta plans to increase its optical module budget by 90% in 2025, mainly for 800G and 1.6T module procurement.
4. Rise of Chinese Manufacturers: Innovation and Performance Boom
1) Leading Companies Report Outstanding Results
Eight listed companies, including InnoLight, Accelink, Eoptolink, and HGTech, saw explosive growth in 2024:
InnoLight: Revenue ¥23.86 billion (+122.6%), net profit ¥5.17 billion (+137.9%)
Eoptolink: Revenue ¥8.65 billion (+179%), net profit +312%
Overseas markets are major growth engines. For example:
InnoLight: 86.8% of revenue from overseas
Eoptolink: 78% from overseas
Top 3 global vendors of 800G modules in 2024 are all Chinese:
InnoLight: 40% market share
Eoptolink: 22%
Accelink: 15%
Key features:
Silicon photonics adoption >30% (industry average: 15%)
Overseas revenue >75%
R&D spending rate 4–6% (industry average: 3.2%)
2) Broad Product and Capacity Expansion
These firms offer full portfolios from 400G to 1.6T, including silicon photonics, EML, VCSEL, and thin-film lithium niobate solutions, and are developing both LPO and CPO technologies. InnoLight has tested 3.2T optical engines; Accelink’s 1.6T monthly capacity reaches 500K units; HGTech aims to raise capacity to 1 million/month in Q2—but still can’t meet demand.
3) Global Manufacturing to Mitigate Geopolitical Risk
To bypass U.S. tariffs, companies are building plants in Thailand, Malaysia, and Vietnam, forming a "China design + Southeast Asia manufacturing" model. HGTech, TFC, Liantech, and Cambridge Technology have secured overseas capacity and compliance certifications to maintain competitiveness and market access.
5. Technical Frontiers: CPO, LPO, Smart Modules, and Standards Race
1) CPO: Future of AI Data Center Interconnects
Co-Packaged Optics (CPO) integrates optical engines directly with ASICs, eliminating traditional optical ports, increasing bandwidth density, and reducing power.
First CPO switches expected in Q3 2025, with commercial rollout in 2026
Leaders: Intel, Cisco, Broadcom
TSMC has initiated 3nm silicon photonic CPO module trials, laying the groundwork for 1.6T and beyond
The commercialization path is clear:
2024: OSFP 3.2T optical engine prototype testing (InnoLight)
2025 Q3: First CPO switch launch (Intel/Broadcom)
2026: Mass production
According to Yole, CPO modules will be 30% more expensive than pluggable modules but reduce total cost of ownership (TCO) by 40%.
2) LPO: Cost-Effective Alternative for Mid-Short Distances
Linear-drive Pluggable Optics (LPO) eliminate the DSP chip, reducing cost by ~30% and power by ~20%. Ideal for short- to medium-range data center interconnects, LPO is a viable alternative to CPO.
3) Smart Modules and Autonomous Operations
Future modules will integrate temperature sensing, power monitoring, and fault prediction. Combined with AI algorithms, they enable self-driving networks, improving data center energy efficiency by around 15%.
4) Intensifying Standards Battle
COBO (international) and CAICT (China) are leading the development of CPO-related standards. Gaining influence in standards will be strategically crucial for shaping the ecosystem and market dominance.
6. Conclusion
Powered by the dual engines of AI and cloud computing, the optical module industry is evolving from a support role into strategic infrastructure. Whether it’s 800G/1.6T modules for core data centers or high-density deployments at the edge, demand is exploding across the board.