23Jun, 2026

GIGALIGHT Releases White Paper on Networking Technologies for Tool Intelligence and National Computing Networks

June 23, 2026, Shenzhen, China. – GIGALIGHT recently announced the release of its white paper dedicated to Networking Technologies for Tool Intelligence and National Computing Networks. This white paper presents GIGALIGHT’s forward-looking technical positioning and industry outlook, serving as an internal guideline for the company’s differentiated product and technology R&D. Most products and technologies covered in the document have obtained patent authorizations from the National Intellectual Property Administration of China.

The outline of GIGALIGHT’s White Paper on Networking Technologies for Tool Intelligence and National Computing Networks is listed below:

I.Building Low-Latency and Low-Power Data Center Interconnect Networks with NRZ Modulation and Half-DSP Technologies
1.NRZ Modulation: Mature and Reliable for Ultra-Low Latency Performance
2.Half-DSP Modulation: LPO and Hybrid Optical Architectures

II.O-Band Non-Coherent DWDM as a Complement to Coherent Communication for Medium-Reach DCI

III.Three Flexible 1.6T Optical Interconnect Architectures for AI and Data Center Applications
1.Standard 1.6T Pluggable Optical Transceivers
2.External Light Source 1.6T Pluggable Optical Module
3.External Light Source 1.6T (16×100G) NPO Solution Near-Package Optics (NPO)

IV.Ultra-Low-Latency All-Optical Switching at Key Network Nodes Using AWGR-Based Architectures
1.Arrayed Waveguide Grating Router (AWGR) Technology
2.AI Cluster Architecture with AWGR + EDFA
3.AI Cluster Architecture with AWGR + SOA

Full text can be accessed via GIGALIGHT’s official website . Comments and suggestions from industry peers are highly welcome.

About GIGALIGHT

As an open optical networking explorer, GIGALIGHT integrates the design, manufacturing, and sales of both active and passive optical devices and subsystems. The company’s product portfolio includes III-V optical modules, silicon photonics modules and silicon-based NPO/CPO engines, liquid-cooled optical modules, passive optical components, Active Optical Cables (AOCs), Direct Attach Cables (DACs), coherent optical communication modules, OPEN DCI BOX subsystems based on coherent and O-band DWDM optical modules, and UHD SDI video optical transceivers. GIGALIGHT focuses on applications including AI data centers, 5G transport networks, metropolitan WDM transmission, and ultra-HD broadcast and video, positioning itself as an innovative designer of high-speed optical interconnect hardware solutions.

23Jun, 2026

GIGALIGHT White Paper: Networking Technologies for Tool Intelligence and National Computing Networks

Computing power is a tool rather than intelligence. Alternatively, we ought to redefine a new type of tool intelligence. Aristotle’s Organon was not composed for everyday utensils. It establishes definitions of categories and logic. Why are categories and logic vital to all human undertakings? Because categories and logic constitute reflection upon thought itself. Adopting a tool-oriented mindset allows us to navigate toward sound scientific and philosophical trajectories.

The national computing power grid demands clear technical insights. Such insights must support upgradability and cross-technology compatibility. To prevent sound computing infrastructure from quickly becoming obsolete amid technological iterations, we must endow such infrastructure with inherent advancement. This advancement belongs to the unified, indivisible category. If we merely follow conventional, non-innovative development paths, the computing power grid will evolve into a consumptive, bloated architecture. Such a framework cannot guide us toward historically rigorous modes of thinking and risks stagnating subsequent development work. Since we define the computing power grid as embodying a degree of tool intelligence, we must operate and unleash this intelligence at minimal cost.

When a person lacks sufficient rest, their mind grows sluggish and briefly stalls or loses vitality. By analogy, the computing power grid resembles the human brain: it must adopt a lightweight daily architecture built for low latency and low-cost innovation, while boasting inherent competitive edges over networking systems deployed in the United States.

I. Building Low-Latency and Low-Power Data Center Interconnect Networks with NRZ Modulation and Half-DSP Technologies

Within AI computing data center networks, NRZ (Non-Return-to-Zero) modulation and “Half-DSP” / linear-drive solutions, including LPO (Linear Pluggable Optics) and Hybrid architectures, have become important technology approaches for achieving both low latency and low power consumption.

These two technology paths are designed for different data rates and transmission distance scenarios. The core objective is to meet the requirements of high-speed interconnects while addressing the high power consumption and latency challenges associated with traditional full-DSP optical modules.

1. NRZ Modulati2. “Half-DSP” Modulation Solutions: LPO and Hybrid Technologiese, Reliable, and Optimized for Ultra-Low Latency

NRZ is a traditional binary modulation technology where each symbol carries one bit of information. In short-distance interconnect applications, its simple physical-layer characteristics make it a preferred solution for latency-sensitive networks.

Application Scenarios

NRZ is mainly used in interconnect scenarios with single-channel speeds ≤200Gbps.

With continued technology evolution, some high-performance SerDes solutions are exploring NRZ-based optimization approaches for higher-speed applications. However, mainstream high-speed optical interconnects at 400G/800G and beyond have transitioned to PAM4 modulation.

Key Advantages

Ultra-Low Latency
NRZ does not require complex digital signal processing (DSP) for signal equalization and error correction. The electrical-to-optical conversion process introduces almost no additional processing latency.

Low Power Consumption
With a simpler circuit architecture and without the high power overhead of DSP chips, NRZ-based optical modules typically maintain low power consumption levels, generally within several watts.

High Signal-to-Noise Ratio Tolerance
In optical fiber transmission, channel loss is relatively low. The two-level NRZ signal provides strong resistance to interference and maintains good signal integrity.

Limitations

NRZ has relatively low spectral efficiency and cannot efficiently support high-speed transmission beyond 50Gbps per wavelength.
Therefore, in mainstream 400G/800G and higher-speed data center and AI computing interconnect architectures, NRZ is gradually being replaced by PAM4 modulation.

GIGALIGHT’s 200G QSFP-DD S8/PSM8 parallel optical module represents a milestone product in this field.

2. “Half-DSP” Modulation Solutions: LPO and Hybrid Technologies

“Half-DSP” is not a formal industry standard term. It generally refers to reduced-DSP or DSP-free optical architectures, including:

LPO (Linear Pluggable Optics)
Hybrid / LRO (Linear Receive Only)

These solutions aim to maintain sufficient signal integrity while significantly reducing the power consumption and latency associated with traditional full-DSP optical modules.

LPO (Linear Pluggable Optics)

The LPO architecture completely removes the DSP chip from the optical module and retains only the linear driver and TIA (Transimpedance Amplifier).Signal processing and equalization are mainly handled by the advanced equalization capability of the switch-side ASIC.

Hybrid (Half Retimed & Half Linear) / LRO (Linear Receive Only)

Hybrid solutions typically retain partial linear-drive functionality at the transmitter or receiver side, or utilize simplified DSP logic in specific signal paths to achieve a balance between performance and power efficiency.

Key Characteristics

Balanced Power Efficiency and Performance
Compared with LPO, Hybrid solutions can provide improved signal integrity for longer transmission distances (such as 100m–2km) while still achieving approximately 20%–30% power savings compared with full-DSP solutions.

Latency Optimization
Although some signal processing capability is retained, simplified algorithms or reduced DSP functions enable significant latency reduction.

Application Scenarios

Hybrid and LRO solutions are suitable for 100m–2km short-to-medium distance interconnect applications, where transmission distance requirements remain important while power efficiency is highly valued.

Core Technology Comparison

GIGALIGHT’s Half-DSP / Hybrid optical modules and active optical cable (AOC) product portfolio represent engineering innovations in this field, providing practical solutions for large-scale network deployment with enhanced feasibility compared with traditional LPO/LRO approaches.

Representative products include:

HYBRID Single-Mode Optical Modules

HYBRID Multimode and Single-Mode AOC

HYBRID Copper Cable Solutions

II. O-BAND Non-Coherent DWDM as an Efficient Complement to Coherent Communication for Low-Latency and Cost-Effective Medium-Distance DCI

With the rapid growth of AI computing clusters and distributed data centers, Data Center Interconnect (DCI) networks are facing increasing demands for 400G/800G/1.6T high-speed optical interconnect solutions.

Although traditional coherent communication technologies provide excellent long-distance transmission capabilities, their complex DSP architecture, high power consumption, higher cost, and increased transmission latency have created efficiency and cost challenges for short- and medium-distance DCI applications within 30km.

To address this challenge, GIGALIGHT has introduced an O-BAND (1310nm window) non-coherent DWDM product portfolio.

By leveraging the natural zero-dispersion characteristics of the O-BAND window, this architecture eliminates the need for DCM (Dispersion Compensation Module) and enables an efficient transmission solution optimized for medium-distance data center interconnect applications.

Compared with traditional C-BAND coherent solutions, O-BAND non-coherent DWDM provides the following key advantages:

Low Latency: By eliminating complex DSP-based dispersion compensation processing, the solution significantly reduces link transmission latency, making it highly suitable for latency-sensitive applications such as AI training and Real-time computing resource scheduling .

Low Power Consumption: By adopting Silicon Photonics integration technology and PAM4 modulation architecture, the overall power consumption is significantly lower than traditional coherent optical modules.

Cost Efficiency: The architecture eliminates the need for expensive coherent components, DCM modules, and complex optical line system designs, significantly reducing: CAPEX, Network deployment cost and Operation and maintenance expenses.

High-Density Deployment: The solution supports high-density optical module form factors including QSFP-DD and OSFP,meeting the requirements of next-generation large-scale data center deployments.

Open and Flexible Architecture: The solution supports open optical networking architectures and is compatible with white-box switches , open optical line systems and Improving overall network flexibility and deployment efficiency.

From 100G to 800G and further toward 1.6T, GIGALIGHT has established one of the industry’s most complete O-BAND DWDM pluggable product portfolios, covering different generations and capacity requirements for data center interconnect networks.

Product Portfolio:

1. 100G QSFP28 DWDM1

Single wavelength 100G transmission
Simplified deployment and operation
System supports up to 16-channel wavelength transmission

2. 200G/400G QSFP-DD DWDM4

4-wavelength aggregation
50G/100G per wavelength
System supports up to 16-channel wavelength transmission

3. 800G OSFP/QSFP-DD 2×PSM DWDM4 (O-BAND)

Single wavelength 100G PAM4
System supports up to 16-channel wavelength transmission

4. 1.6T OSFP224-EL 2×PSM DWDM4 (O-BAND)

Single wavelength 200G PAM4
External light source architecture
System supports up to 16-channel wavelength transmission

O-BAND DCI Interconnect Application Scenario

Transmission Distance: 2km–30km

III. Three Flexible Solutions for 1.6T AI & Data Center Interconnect

With the explosive growth of AI computing requirements, data center interconnect bandwidth is rapidly evolving from 400G/800G toward 1.6T.

To meet deployment requirements across different application scenarios, the industry is developing three major technology approaches for 1.6T optical interconnects. Each solution provides different advantages and is optimized for different requirements in terms of cost structure ,power budget and deployment scalability.

1. Standard 1.6T Pluggable Optical Transceiver

Standard 1.6T OSFP224 Optical Module

The 1.6T OSFP224 optical module is a next-generation high-speed optical transceiver featuring:

Total bandwidth of 1.6Tbps
OSFP224 form factor
Designed to support AI clusters and high-performance computing bandwidth requirements

It represents one of the mainstream solutions for next-generation AI networking deployments.

Core Parameters and Definition

Data Rate and Form Factor

Total Data Rate : 1.6Tbps total bandwidth is achieved through 8 × 200G PAM4 electrical channels. After protocol overhead, the effective payload reaches approximately 1.6Tbps.

OSFP224 Form Factor：OSFP224 is an evolution of the traditional OSFP form factor. It features:

224 electrical contacts
Support for 16 electrical lanes
Higher port density
Enhanced thermal capability

It is designed for high-density AI switch deployments.

Standard 1.6T Pluggable Module Solutions

2.External Light Source 1.6T Pluggable Optical Module (External Light Source Pluggable, ELS-Pluggable)

This architecture separates the laser source from the 1.6T OSFP224 optical module. A centralized:

Rack-level external light source
Board-level external light source
Co-packaged light source

provides optical power to multiple pluggable modules through fiber coupling.

The optical module itself only retains modulators and photodetectors and required electrical processing components. This architecture significantly reduces thermal load at the individual module level.

As one of the pioneers proposing the External Light Source 1.6T Pluggable Optical Module architecture, GIGALIGHT provides the following solutions:

3. External Light Source 1.6T (16×100G) NPO Solution Near-Package Optics (NPO)

The NPO architecture represents a transition technology between pluggable optics and Co-Packaged Optics (CPO). In the NPO architecture, the optical engine is no longer integrated inside the switch ASIC package.

Instead, it is installed immediately outside the package：Optical engine distance from ASIC: <50mm

Combined with an external light source, NPO enables ultra-short-distance high-speed interconnects through Silicon Photonics or Polymer waveguide technologies.

The 1.6T NPO solution is typically deeply integrated with Silicon Photonics platforms to achieve extremely low electrical-to-optical transmission loss.

It should be noted that practical NPO optical engine architectures generally need to be based on:

16×100G physical architecture
32×100G physical architecture

This represents the key differentiation point compared with 1.6T pluggable optical modules.

Although the industry is exploring 3.2T NPO and 6.4T NPO

Architectures based on 224G SerDes technology, related standards have not yet been finalized.

Comparison of Three 1.6T Solutions

Deployment Roadmap

The three solutions are not mutually exclusive alternatives.

Instead, they represent a layered deployment strategy targeting different technology maturity stages and application scenarios.

Near Term (2024–2026): Standard pluggable optical modules will remain the mainstream solution, supporting smooth upgrades for most data center deployments.

Mid Term (2027+): External light source pluggable architectures are expected to achieve rapid adoption in large-scale AI clusters by balancing deployment flexibility and power optimization

Long Term (2027+): With the maturity of Silicon Photonics technology and standardization progress, NPO solutions are expected to become a major architecture for AI supercomputing interconnects, providing the foundation for future 3.2T and 6.4T network evolution.

IV. Enabling Ultra-Low Latency All-Optical Switching at Critical Network Nodes with AWGR + EDFA / AWGR + SOA Architectures

1. AWGR (Arrayed Waveguide Grating Router)

AWGR is a passive wavelength routing engine that utilizes its cyclic wavelength routing characteristics to deterministically map different optical wavelengths within a WDM signal to corresponding output ports.

Unlike traditional electrical switching architectures, AWGR does not require lookup tables, packet scheduling and optical buffering.

Therefore, it completely eliminates the traditional optical-electrical-optical (O-E-O) conversion process in electrical switching systems.

The optical signal remains entirely in the optical domain throughout the transmission process.

A single-stage optical switching operation introduces only nanosecond-level latency, which is:

2–3 orders of magnitude lower than traditional OEO architectures
Highly deterministic with extremely low latency jitter

This capability makes AWGR-based optical switching highly suitable for:

AI computing clusters
Low-latency computing networks
High-performance computing scheduling applications

2. AI Computing Cluster AWGR + EDFA Optical Switching Topology

1) Architecture Overview

This architecture is mainly designed for data center inter-rack optical switching and cross-campus long-distance all-optical switching hubs.The topology adopts a three-level flattened architecture:

Compute Access Layer → Core Wavelength Switching Layer → Optical Amplification Output Layer

Access Layer

Each AI computing node / GPU server outputs optical signals through:
100G PSM DWDM4 (C-BAND) optical modules
Each optical port directly carries:
Four fixed C-band wavelengths
The optical signals are then aggregated through a:
DWDM multiplexer (MUX)

Core Switching Layer

After wavelength multiplexing, signals are transmitted into an:
N×N AWGR wavelength router
The AWGR performs optical routing switching based on wavelength assignment.
The switching process is completed entirely in the optical domain.

Optical Compensation Output Layer

After AWGR switching, the system experiences approximately:
8dB inherent insertion loss
An EDFA (Erbium-Doped Fiber Amplifier) is used to compensate for this optical loss.
The amplified signal is then transmitted to:
Next-stage switching nodes
Target computing clusters
Supporting transmission distances at the:
10km level

2) Topology Characteristics

Key advantages include:

No multi-stage O-E-O conversion
Core switching latency below 1μs
Simplified network architecture
Suitable for large-scale computing hubs with relatively stable network topology

Technical Parameter Comparison

3. AI Computing Cluster AWGR + SOA Optical Switching Topology

This architecture is mainly designed for:

In-rack optical switching
Same data center high-density optical switching hubs
Ultra-low latency AI cluster interconnects

The topology adopts a compact architecture combining:Active Scheduling + All-Optical Switching + Integrated Optical Amplification

Wavelength Scheduling Layer

Each computing node connects to:

O-BAND DWDM optical components
O-BAND multi-wavelength tunable optical engines

The scheduling controller dynamically adjusts output wavelengths according to GPU communication requirements, including Gradient synchronization and Parameter broadcasting.

GIGALIGHT provides O-BAND DWDM parallel optical modules with different baud rates to support medium-distance AI computing interconnect applications.

Core Optical Switching Layer

The dynamically scheduled wavelength signals are transmitted into:

AWGR wavelength routers

The AWGR performs:

Wavelength-based optical routing
Nanosecond-level optical path switching

Integrated Optical Amplification Layer

The architecture uses:

SOA (Semiconductor Optical Amplifier) for amplification of weak optical signals after switching.

Compared with EDFA, SOA provides:

Smaller form factor
Better integration capability
Suitability for high-density rack deployment

SOA can also assist with:

Wavelength conversion
Enhanced scheduling flexibility

Topology Characteristics

The complete switching process latency can be reduced to:

Below 100 nanoseconds
The power consumption of a single optical switching chip is approximately:
1/10 of a comparable electrical switching ASIC

This architecture is highly suitable for:

Large-scale AI training clusters
Thousands of GPU interconnect environments
Low-latency east-west GPU communication

The network architecture diagram

Key Node Selection Recommendations

Conclusion

The core essence of a computing power network resides in its capability to invoke tools. A pure computing power network, an expensive piece of infrastructure, would be utterly meaningless without a society-wide foundation of lifelong learning.

From any perspective of commercial development, blind dependence on large computing models will clash with knowledge undertakings constructed through language. Commercial organizations may only conduct innovative thinking based on computing power networks; they must not regard computing power itself as a competitive advantage.

Computing power networks should function as networks that boost the intellectual capacity of all citizens. Otherwise, trillions of capital invested will fail to deliver any positive outcomes.

Just as traditional telecommunication pipelines were built to connect information, computing power networks exist to connect humans with instrumental intelligence.

Only when we precisely define the core value and positioning of computing power networks can this network become a digital infrastructure worthy of massive national investment.

12Jan, 2026

The Parallel Evolution Logic and Industry Choices of HYBRID, LPO, LRO, and DSP— Why HYBRID Is More Practical Than LRO from a System Engineering Perspective

I. Introduction: Parallel Coexistence, Not a Battle of Routes

As AI and data center interconnects rapidly advance toward 800G and 1.6T, LPO, LRO, and traditional DSP-based optical modules are not engaged in a “winner-takes-all” replacement battle. Instead, constrained by different system requirements, transmission distances, and levels of industry maturity, they form a long-term pattern of parallel coexistence.

Within this parallel framework, HYBRID (semi-DSP) solutions—representing a more engineering-oriented intermediate architecture—are demonstrating clear advantages over typical LRO solutions in terms of overall system and industrial value.

Based on a systematic review of the parallel evolution logic of LPO, LRO, and DSP, and combined with real-world 800G HYBRID product implementations, this paper compares the strengths and weaknesses of each approach and focuses on a key question:

Why is HYBRID more practical and advantageous than LRO in real-world deployment?

II. Review of the Three Fundamental Architectures: DSP, LPO, and LRO

1. DSP-Based Modules: The “Ballast Stone” of Performance and Ecosystem

(1)Architecture Characteristics：Full DSP on both TX and RX sides

(2)Advantages

Strongest signal processing capability (equalization, CDR, FEC, nonlinear compensation)
Supports medium, long, and ultra-long reach transmission
Mature industry ecosystem, complete standards, true plug-and-play

(3)Disadvantages

High power consumption (800G typically >14–16 W)
Higher latency
Highest cost

(4)Typical Applications

Metro networks, backbone networks, DCI
Mission-critical links with extreme reliability requirements

2. LPO: Ultimate Power Efficiency with the Deepest System Coupling

(1)Architecture Characteristics : No DSP inside the module . Signal processing fully shifted to the host SerDes

(2)Advantages

Lowest power consumption (30–50% reduction compared to DSP)
Ultra-low latency
Lowest module BOM cost

(3)Disadvantages

Extremely stringent requirements on host SerDes and channel consistency
Limited transmission distance (typically ≤100m)
Complex system tuning and immature ecosystem

(4)Typical Applications

Intra-rack and inter-rack ultra-short-reach interconnects in AI clusters

3. LRO: A Conceptual Compromise

(1)Architecture Characteristics : DSP retained on TX side,Linear RX architecture

(2)Advantages

Lower power consumption than full DSP
Better reach capability than LPO

(3)Practical Challenges

Highly fragmented TX simplex DSP implementations
Poor chip reusability and weak economies of scale
Ecosystem has not reached mainstream adoption

III. HYBRID: A More Engineering-Mature “Semi-DSP” Path

1. The Essential Definition of HYBRID

HYBRID is not simply equivalent to LRO. Instead, it is a system-level DSP resource reallocation methodology:

HYBRID = Only half of the TX/RX paths inside the module pass through DSP, and for the module-level transmit/receive links, only TX or RX passes through DSP,while the remaining paths adopt linear architectures.

Conceptually, this can be understood as: HYBRID ≈ (LRO + LTO) / 2

Below is the patented HYBRID architecture description from GIGALIGHT:

HYBRID 800G Multimode Architecture Diagram (Patent Applied)

HYBRID 800G Single-Mode Architecture Diagram (Patent Applied)

HYBRID 800G Copper Cable Architecture Diagram (Patent Applied)

2. Key Differences: HYBRID vs. LRO

Dimension	LRO	HYBRID
DSP Type	TX simplex DSP	Mature duplex DSP
Chip Reusability	Very low	High, reuse of existing DSP
Market Scale	Niche, customized	Scalable, mass-producible
Supply Chain Risk	High	Low
System Consistency	Medium, one-side linear	Medium, one-side linear

Core Conclusion:
HYBRID does not introduce a new class of highly specialized DSP chips. Instead, it reuses mature duplex DSP architectures and simply reduces the number of active channels.

Furthermore, early system-level validation shows that certain LTO-based system links can even outperform LRO links, reinforcing HYBRID’s clear advantage over LRO in real industrial deployment.

IV. Comprehensive Advantages and Practical Challenges of HYBRID Key Advantages of HYBRID

1. Key Advantages of HYBRID

Significant power reduction：~20–30% lower than full DSP solutions
Ultra-low link latency：DSP count reduced by half; latency comparable to LRO
Controllable signal quality：MMF 50m: PRE-FEC BER up to E-7 / E-8，SMF 500m: PRE-FEC BER up to E-10
Clear cost optimization：~20% cost reduction versus traditional DSP solutions
Support for higher channel density：Viable path toward 16-channel / 3.2T pluggable modules

2. Limitations and Challenges of HYBRID

Non-DSP RX paths place slightly higher requirements on host SI tuning
Requires system-level co-optimization rather than pure “plug-and-play”
Still in early stages of large-scale deployment, requiring customer-side collaboration

However, compared with LPO and LRO, the engineering and deployment risks of HYBRID are significantly more controllable.

V. The Inevitability of Parallel Coexistence: Why There Is No Single Winning Route

Scenario	Optimal Solution
≤100m, ultra-low latency	LPO
100m – 2km, power/performance balance	HYBRID / LRO
≥2km, maximum reliability	DSP

These three solutions address different system constraints and optimization targets, rather than forming a simple generational replacement relationship.

VI. Conclusion: HYBRID Is the Most Realistic Intermediate Solution

LPO is an idealized extreme solution
DSP is an irreplaceable foundational solution
HYBRID is currently the most engineering-feasible intermediate solution

Compared with LRO, HYBRID’s decisive advantage lies in the fact that:
It does not rely on highly specialized TX simplex DSPs with extremely limited market demand, but instead reuses mature, scalable duplex DSP architectures.

This enables HYBRID to achieve clear overall competitiveness in power consumption, cost, ecosystem maturity, supply chain stability, and system-level deployment.

In the foreseeable future, LPO, HYBRID, and DSP will continue to coexist, jointly forming the technological foundation of AI and data center interconnects.

About GIGALIGHT

As an open optical networking explorer, GIGALIGHT integrates the design, manufacturing, and sales of both active and passive optical devices and subsystems. The company’s product portfolio includes optical modules, silicon photonics modules, liquid-cooled modules, passive optical components, active optical cables, direct attach copper cables, coherent optical communication modules, and OPEN DCI BOX subsystems. GIGALIGHT focuses on serving applications such as data centers, 5G transport networks, metropolitan WDM transmission, ultra-HD broadcast and video, and more. It stands as an innovative designer of high-speed optical interconnect hardware solutions.

10Nov, 2025

Breakthrough Development of CPO in the AI Era

When ChatGPT’s daily API calls exceeded 1 billion and AI large model training consumed hundreds of billions of floating-point operations in a single run, computing power has become the core production factor in the digital economy era. Supporting this computing power tsunami is not only the “computing heart” of GPU clusters but also the “transmission vessels” of optical communication networks. In 2025, optical communication technology centered on CPO (Co-packaged Optics) has reached a triple inflection point of “technological breakthrough – demand explosion – performance realization.” The aggressive layout of giants like NVIDIA and Broadcom, along with accelerated breakthroughs in domestic industrial chains, is driving CPO to become the core force reshaping AI data centers and leading a new round of technological transformation.

Technical Breakthrough: From Architectural Reconstruction to Performance Leap

CPO is not a local optimization of traditional optical modules but achieves a fundamental transformation of optical interconnect links through “optoelectronic co-packaging.” Its core logic is the deep integration of optical engines with switching chips/computing chips at the packaging level, reconstructing the traditional architecture of “switching chip → PCB traces → optical module → optical fiber” into a new architecture of “chip → millimeter-level electrical interconnect → optical engine → optical fiber,” fundamentally breaking through physical transmission bottlenecks.

The 2025 NVIDIA GTC Conference became the critical node for CPO technology explosion. The conference explicitly announced that the GB300 chip and next-generation Rubin platform to be launched in the second half of the year will fully adopt CPO technology, proposing three mandatory metrics: transmission energy efficiency ratio must be lower than 8pJ/bit (a 47% reduction compared to existing solutions), 1U equipment bandwidth density must be no less than 100Gbps/mm, and end-to-end latency must be controlled within 0.5 microseconds. This requirement directly sentences traditional pluggable optical modules to “death”—800G pluggable modules have power consumption of up to 15-20W and bandwidth density of only 30Gbps/mm, completely unable to meet the needs of AI computing clusters. Through CPO technology, the Rubin platform increases the data transfer rate between chips to 6.4Tbps, with interconnect quantities exceeding 256, enabling AI cluster computing power density to increase by 3.5 times.

Technological innovation continues to iterate, and cost control has achieved key breakthroughs. NVIDIA’s research team proposed an innovative solution using single-mode fiber (SMF) instead of traditional polarization-maintaining fiber (PMF), solving the polarization state drift problem introduced by SMF coupling by integrating an active polarization tracking system on the silicon photonic chip. Experimental data shows this solution can reduce packaging costs by approximately 2.7 times while only introducing 0.74dB additional insertion loss and 0.17pJ/b energy efficiency loss, clearing cost barriers for CPO large-scale application. Additionally, CPO technology achieves significant power optimization by eliminating complex signal compensation circuits

Industrial Chain Reshaping: Global Competition and Domestic Breakthrough

The commercialization of CPO technology is not only a product upgrade but also a comprehensive reshaping of the optical communication industrial chain. In the traditional optical module industrial chain, the packaging assembly link occupies over 60% of the value share, while CPO shifts the value center to high-barrier core technology links, forming a new value chain of “upstream core devices – midstream advanced packaging – downstream system integration,” presenting an obvious “pyramid” value distribution pattern.

In the upstream core device field, domestic substitution is accelerating breakthroughs. Silicon photonic chips, as the “brain” of CPO, determine the performance ceiling.

The midstream packaging link is welcoming a process revolution. CPO packaging adopts a 2.5D integrated architecture of “chip – silicon interposer – optical engine,” with interconnect density reaching 10^4 per mm², and process complexity 3 times higher than traditional optical modules.

Global giants are accelerating ecological layout. Broadcom launched the 51.2T Bailly switch, which has achieved 1 million 400G port failure-free operation in Meta data centers; AMD invested NT$8.64 billion to build a silicon photonics R&D center in Taiwan; TSMC and Broadcom are testing 3nm MRM CPO, planning mass production in 2026; NVIDIA invested approximately $120 million to build an AI data center in Munich, Germany, which will deploy 10,000 Blackwell GPUs supporting CPO.

Commercial Implementation: From Verification to Large-Scale Deployment

2025 has become the critical year for CPO technology to move from the laboratory to commercial use, with AI data centers becoming the core application scenario. NVIDIA Spectrum-X CPO switches have been adopted by Oracle and Meta, enabling system energy efficiency to increase by 3.5 times, reliability to improve 10 times, and deployment time to shorten to 77% of the original. Broadcom and Tencent’s cooperative 25.6Tbps CPO switching equipment, through coupling and packaging 4 3.2Tbps optical engines with switching chips, achieved a breakthrough of 64 optical ports in 1U equipment, with bandwidth density reaching 2.8 times that of traditional solutions.

In AI training scenarios, CPO technology’s low-latency advantage is particularly critical. By eliminating signal delay from long-distance electrical connections, CPO increases GPU cluster computing power utilization by over 20%, potentially reducing GPT-6 large model training cycles from 18 months to 12 months. Tech giants like Microsoft and Google have significantly increased CPO procurement.

Market size shows explosive growth momentum. The global CPO market size is expected to reach $8.6 billion in 2025 and exceed $50 billion in 2027, with an annual compound growth rate as high as 37-40%. From the penetration rhythm perspective, by the end of 2025 it is in the early commercial stage (AI data center penetration <10%), entering mass production peak in 2026 (penetration 20-30%), achieving large-scale application in 2027 (>40%, HPC accounting for half), and by 2033 the market size is expected to reach $250 billion.

Future Trends: Technology Evolution and Ecosystem Integration

CPO technology’s evolution path is clear and explicit, upgrading from the current 1.6T rate to 3.2T and 6.4T, transitioning from LPO to pure CPO and then to future optoelectronic integrated (OIO), with continuously improving integration being the core thread. LightCounting data shows the global LPO market size will reach $1.8 billion in 2025, but by 2027 its proportion will drop from 25% to 8%, being rapidly replaced by CPO technology.

Technical challenges still require continuous breakthroughs. Current CPO faces issues such as integration complexity, difficult thermal management, and dependence on imported high-end materials. Future innovations are needed in silicon photonic chip integration, passive alignment accuracy, liquid cooling efficiency, etc. With the popularization of 3.2T and 6.4T rates, the integration of optical engines will become increasingly high, and an integrated solution of “optical engine + chip” may emerge, placing higher demands on corporate R&D capabilities.

Industrial chain ecology will accelerate integration. CPO technology involves multiple fields including chip design, optical devices, advanced packaging, and equipment manufacturing. No single enterprise can cover the entire chain, making cross-border cooperation an inevitable trend. Domestic enterprises are building competitive advantages through models such as “chip – complete machine” binding and vertical integration.

The outbreak of CPO technology is not only a revolution in the optical communication industry but also an upgrade and reconstruction of digital economy infrastructure. Against the backdrop of continuous AI computing power demand explosion, CPO, with its core advantages of ultra-high bandwidth, ultra-low latency, and ultimate energy efficiency, is becoming the key support to break through computing power transmission bottlenecks. With continuous investment from global giants and accelerated breakthroughs in domestic industrial chains, CPO will move from technological innovation to large-scale popularization, bringing profound changes to AI factories, supercomputing centers, cloud computing and other fields, and leading human society into a more efficient digital era.

25Jun, 2025

The Development Prospects of Optical Transceivers in the AI Era

The explosive growth of artificial intelligence (AI) technology is reshaping the underlying logic of the global digital infrastructure. As the core transmission carrier of data centers and communication networks, the optical transceiver industry is standing in the “eye of the storm” of the AI computing revolution. From the rate iteration of 800G to 1.6T, from the technological leap from traditional pluggable to CPO (co-packaged optics), optical transceivers are not only the “highways” of data transmission, but also the “blood transfusion channels” for AI large model training and reasoning. This article will deeply analyze the technological evolution, market demand and competitive landscape of the optical transceiver industry in the AI era, and look forward to its future development prospects.

AI Computing Power Demand: The Underlying Logic Driving the Iteration of Optical Transceiver Technology

Exponential Demand for Bandwidth in Large-Model Training

The parameter count of individual AI large models has surpassed the trillion level (e.g., GPT-4), and the computing power required for training doubles every 3-4 months, far outpacing Moore’s Law. This directly drives the upgrade of optical interconnection bandwidth within data centers from 100G to 800G/1.6T. Taking NVIDIA’s DGX H100 server cluster as an example, its internal interconnection requires a total bandwidth of at least 4.8 Tbps, with the demand for optical transceivers per single cabinet exceeding 500 units.

Technical Challenges of Low Latency and High Energy Efficiency

AI computing has extremely low tolerance for data transmission latency (requiring less than 100 nanoseconds), while the power consumption of traditional pluggable optical transceivers already accounts for over 30% of the total energy consumption of data centers. CPO technology, by directly packaging the optical engine with the ASIC chip, can reduce power consumption by 40% and latency by 50%, becoming the next technological high ground. According to LightCounting’s forecast, the CPO market size will exceed $5 billion by 2027.

Commercial Breakthroughs in Silicon Photonics Technology

Silicon photonics technology integrates lasers, modulators, and detectors through CMOS processes, significantly reducing the cost and size of optical transceivers. Companies like Intel and Cisco have already launched silicon photonics 400G transceivers, and domestic manufacturers are also accelerating mass production. Yole predicts that the market share of silicon photonics modules will increase from 20% in 2023 to over 60% by 2030.

Market Landscape: Global Supply Chain Restructuring and Domestic Opportunities

“Arms Race” Among North American Cloud Providers

Tech giants such as Meta, Microsoft, and Google plan to increase the penetration rate of optical transceivers in AI data centers to 80% by 2024-2025. Taking Microsoft Azure as an example, its planned deployment of 1.6T optical transceivers will reach 2 million units by 2025, with a market size exceeding $3 billion. North American cloud providers are deeply binding with leading Chinese enterprises through the JDM (Joint Design Manufacturing) model.

Technological Breakthroughs by Chinese Manufacturers

With cost advantages and rapid iteration capabilities, Chinese optical transceiver enterprises have captured over 60% of the global market share. According to Omdia data, Chinese enterprises accounted for over 70% of the 800G market share in 2023.

Supply Chain Resilience Amid Geopolitical Tensions

U.S. semiconductor export controls on China have compelled domestic industry chain vertical integration. From optical chips, packaging materials to testing equipment, the domestic substitution rate has increased to over 50%. Meanwhile, leading companies have evaded tariff risks through overseas bases in Thailand and Malaysia, forming a global production capacity network of “Chinese design + Southeast Asian manufacturing”.

Future Trends: From Technological Evolution to Ecosystem Competition

The Rise of LPO (Linear Direct Drive) Technology

Compared with CPO, the LPO solution eliminates the need for DSP chips, reducing power consumption by 20% and cost by 30%, becoming an alternative for medium- and short-distance transmission.

Optoelectronic Integration and Intelligent Operation and Maintenance

Optical transceivers are evolving towards intelligence, integrating functions such as temperature sensing, power consumption monitoring, and failure prediction. The concept of “autonomous driving networks” proposes dynamically optimizing the working state of optical transceivers through AI algorithms, improving data center energy efficiency by 15%.

Standards Competition and Ecosystem Positioning

Industry alliances have become new battlefields for competition. International organizations such as COBO (Co-Packaged Optics Alliance) lead the formulation of CPO standards, while the China Academy of Information and Communications Technology (CAICT) has taken the lead in establishing the “Optoelectronic Integration Technology Committee” to promote the implementation of domestic standards. In the next five years, whoever holds the discursive power in technological standards will dominate the trillion-dollar AI computing power market.

Challenges and Risks

Uncertainty in Technological Iteration

The parallel development of multiple technological routes such as CPO, LPO, and silicon photonics may lead to risks of redundant investment. For example, if the heat dissipation and yield issues of CPO cannot be overcome, it may trigger industrial turbulence due to a shift in technological routes.

The “Bullwhip Effect” in the Supply Chain

Fluctuations in AI computing power demand may lead to inventory backlogs of optical transceivers. In Q4 2023, order cancellations by North American cloud providers caused some enterprises’ inventory turnover days to soar to over 120 days, exposing the fragility of demand forecasting.

The “Gray Rhino” of Geopolitics

The United States may include optical transceivers in the export control list or even restrict the supply of advanced optical chips to China. If domestic 25G and above DFB lasers cannot make breakthroughs, it will hinder the development of high-end products.

Conclusion

Under the dual revolutions of AI and computing power, the optical transceiver industry has evolved from a “supporting role” in the communications field to a strategic infrastructure for the digital economy. In the short term, the mass production of 800G and the R&D of 1.6T will spawn a new round of growth cycles; in the long term, optoelectronic integration and standards competition will define the industry’s final outcome. For Chinese enterprises, only by achieving a balance between technological innovation, supply chain resilience, and adaptation to international rules can they remain invincible in this global competition and cooperation. The future optical transceiver industry is not just a technological contest but a comprehensive game of ecosystems and strategies.

13Aug, 2024

It doesn’t matter where you run, what matters is the right direction

Optical communications field – the main application scenarios of silicon photonic chips

With the rapid development of artificial intelligence (AI) and machine learning (ML) applications, the demand is for components to exchange data quickly and consume as little power as possible while maintaining high computing density, especially in terms of short-distance connections between XPUs (CPU, GPU, and memory).

The AI computing era we are facing has the following characteristics:

Require faster network speeds – rapidly advancing towards 800G/1.6T
Require network products to have lower latency – improve computing efficiency, and increase the response speed and training speed of AI systems
Require network products to have lower power consumption – reduce energy consumption and develop sustainable AI computing power
Require higher reliability of system hardware and software – reduce failure rate, improve availability and user experience of AI services

In this context, traditional pluggable optical modules are already “stretched” in terms of cost performance and power consumption, while highly integrated high-speed silicon photonic chips have become a more superior option due to their obvious advantages in potential price reduction and power consumption. The market’s choice is the best proof. Silicon photonic modules have achieved partial commercial success, and the application of higher-speed silicon photonic modules will also increase significantly.

Silicon photonics is emerging as a promising technology compared to traditional methods. Silicon photonics can enable better communication between computing units such as CPUs and GPUs. Memory units can also be improved to increase the computing power and efficiency of AI applications.

Gigalight Silicon Photonics Transceiver

In response to the comprehensive goal of large-scale commercial use and technological advancement of 800G architecture in AI data centers at home and abroad, Gigalight launched a low-power 800G series silicon photonic modules based on 7nm DSP with a power consumption of 16W, including 800G DR8\DR8+, 800G 2×FR4/2×LR4 data center optical modules, supporting OSFP and QSFP-DD packaging forms, laying a solid delivery foundation for the company to become a pioneer in the field of silicon photonics.

800G OSFP 2×FR4/ 2×LR4

Gigalight’s 800G OSFP 2×FR4 and 800G OSFP 2×LR4 optical module ports use 2× Dual LC interfaces. The transmitter uses a monolithic integrated silicon photonic modulation chip solution, using 4 CWDM4 light sources. The integrated silicon photonic chip realizes 1-to-2 light source and CWDM wavelength division and MZ modulation. The receiver is 2 discrete free-space optical POSAs and adopts a low-power direct-drive 7nm DSP chip solution. The product has the advantages of low power consumption and high performance.

800G DR8/DR8+/DR8++

Gigalight’s 800G OSFP DR8/DR8+/DR8++ optical module port uses a 2×MPO-12/APC interface. The transmitter uses a monolithic integrated silicon photonic modulation chip solution, uses two CWL light sources, and implements 1-to-4 and MZ modulation of the light source in the integrated silicon photonic chip. The receiver uses two discrete FA solutions and a low-power direct-drive 7nm DSP chip solution. This product has the advantages of low power consumption and high performance.

At the same time, Gigalight also provides 800G silicon photonic modules in QSFP-DD package to meet the needs of different users.

800G QSFP-DD DR8/DR8+/DR8++

1310nm CW+MZ+PIN
500m/2km/10km
<14.5W/16W

800G QSFP-DD 2×FR4/ 2×LR4

CWDM4 CW+MZ+PIN
2km/10km
<16W

400G DR4

As early as 2023, Gigalight has mass-produced 400G OSFP-RHS DR4 and 400G QSFP112 DR4 silicon photonic modules. The addition of the latest low-power version 800G silicon photonic series has strongly promoted the company’s confidence in the growth of next-generation optical communication performance. Gigalight has substantially entered the research and development of silicon photonic chips and silicon photonic modules since 2018, and has completed the mass production or sample stage of at least 10 silicon photonic modules. Thanks to the strong collaboration of the global supply chain, more innovative and differentiated silicon photonic modules will be launched in Gigalight in the future.

In the 800G AI computing center, the 800G silicon photonic module can branch two 400G optical modules through single-mode optical fiber to achieve a smooth upgrade of the 400G network.

5Jul, 2024

Silicon photonics is the key to unlocking the full potential of AI

As artificial intelligence (AI) and machine learning (ML) applications grow rapidly, we can foresee that industries and companies will use these systems and tools in their daily processes. As the complexity of these data-intensive applications continues to increase, the need for high-speed transmission and efficient communication between computing units becomes critical.

This demand has sparked interest in optical interconnects, especially for short-distance connections between XPUs (CPUs, GPUs, and memory). Silicon photonics is emerging as a promising technology that can improve performance, cost-effectiveness, and thermal management capabilities compared to traditional approaches, ultimately improving the functionality of AI/ML applications.

Advantages of Silicon Photonics in Artificial Intelligence

Silicon photonics interconnects play a critical and specialized role in managing the growing demand for AI/ML applications across industries. These components need to exchange data quickly and consume as little power as possible while maintaining high computing density. Silicon photonics enables good communication between computing units such as CPUs and GPUs, and memory units can also be improved to increase the computing power and efficiency of AI applications.

Silicon photonics example layout (Source: OpenLight)

Laser integration is essential for generating, modulating, and manipulating optical signals and introducing them into various systems. However, this has always been a major challenge in the field of silicon photonics.

On-chip optical interconnect technology

To meet market demands, companies have begun investing in on-chip optical interconnects to enable scalability from one laser to hundreds of lasers, surpassing the challenges posed by traditional electrical interconnects.

Short-reach optical interconnects using silicon photonics offer a solution that enables high-speed data transmission while reducing power consumption and increasing thermal efficiency (pJ/bit). This is important to reduce heat generation and keep systems running efficiently.

In addition, silicon photonic integration can create smaller and denser photonic integrated circuits (PiCs), facilitating high-density bandwidth connections that are critical to AI/ML workloads. Heterogeneous integration can more efficiently connect lasers and waveguides, resulting in better coupling and reduced power consumption. In addition, with the development of new lasers, thermal efficiency has been improved, and the number of channels and the number of potential wavelengths per channel have been expanded.

Overcoming the challenges of high-density bandwidth connections

In the case of back-end manufacturing, companies can save significant operating and capital expenditures without having to use lasers externally coupled to passive silicon photonic chips. By using more channels per square inch of silicon and combining different active components together, it is possible to use less power and significantly increase the total bandwidth of each PIC.

Gigalight Silicon Photonics 800G Optical Transceiver

Silicon photonics allows for efficient data transmission in AI/ML applications using short-reach optical interconnects. In the case of AI applications such as natural language processing, image recognition, and autonomous driving, large data sets are processed and analyzed in real time.

Fast and efficient data transmission is critical for real-time decision making and optimal system performance. Silicon photonics can provide high-speed data transmission and efficient communication between components, helping to improve the overall efficiency and performance of AI systems in these areas.

By leveraging silicon photonics technology, companies can optimize their AI/ML systems and unleash more powerful computing power for more accurate and responsive results.

The future of silicon photonics in artificial intelligence

The road ahead is full of hope. Silicon photonic technology has the potential to revolutionize artificial intelligence algorithms and further enhance the capabilities of artificial intelligence systems. The use of silicon photonic technology in artificial intelligence can develop smarter systems that handle complex tasks with higher performance and efficiency.

As architects further develop AI networks, silicon photonics and heterogeneous integration will transform the switching layer, replacing traditional packet switching, which will enable lower latency and lower power consumption at the required interconnect density.

It is believed that silicon photonics technology will become a disruptive technology for future AI/ML systems and has significant advantages over traditional electrical signal solutions. This, in turn, could push the boundaries of what is possible in the field of AI.

23Jan, 2024

Exploring the Advantages of 200G (8x25G NRZ) Optical Technology in Data Centers

GIGALIGHT, which has focused on optical communication for eight years, directs your attention to the 200G (8x25G NRZ) technology, delving into its advantages such as low power consumption, low latency, and easy deployment. This provides a deeper understanding of cost-effective optical interconnection solutions within data centers.

There are two technological directions for 200G optical modules: one utilizes the QSFP-DD packaging for an economical 8x25G NRZ network, while the other employs QSFP56 packaging for a 4x50G PAM4 network (to be discussed in a subsequent article).

Compared to PAM4 technology, 200G (8x25G NRZ) boasts advantages in low power consumption, low latency, and easy deployment.

Technical details:

QSFP-DD packaging for 8x25G NRZ network: The 200G optical module uses QSFP-DD packaging, transmitting 25G NRZ signals through 8 channels. NRZ, representing non-return-to-zero signals, involves signal switching between high and low levels. Compared to traditional PAM4 signals, NRZ signals are easier to modulate and demodulate.

Low power consumption: Utilizing 25G NRZ optical components, the module’s power consumption is reduced by 2–3W compared to modules based on PAM4 DSP CDR technology. This offers significant advantages for data center operational costs (OPEX) and cooling efficiency.

Low latency: 200G NRZ excels in special application scenarios with stringent network latency requirements, such as high-frequency trading networks. The transmission method of NRZ signals provides an advantage in terms of latency.

Deployment ease:

Mature industry chain: The technology of 25G NRZ optical chips and components is relatively mature, providing a reliable foundation for the production of 200G NRZ. This makes the development of related 200G switch ports and optical module products relatively easy, facilitating rapid commercial deployment.

Low-cost optical interconnection: Due to mature technology and industry chains, 200G NRZ achieves low-cost optical interconnection within data centers. This offers an economically effective solution for the construction and expansion of large-scale data centers.

200G QSFP-DD SR8

The 200G QSFP-DD SR8 is a multimode parallel product, utilizing an 8-channel 850nm VCSEL array, compliant with the 100GBASE-SR4 protocol standard. Leveraging the advantages of traditional VCSEL platforms, GIGALIGHT has adopted a simple, efficient, and reliable fiber coupling process. By introducing a 45° prism between the laser and the fiber, along with special material treatment on the fiber surface, the fiber coupling efficiency has been increased to over 80%.

200G QSFP-DD PSM8

The 200G QSFP-DD PSM8 is a high-speed product utilizing single-mode parallel technology. In comparison to wavelength division multiplexing (WDM) technology, PSM technology offers cost advantages. GIGALIGHT PSM optical module enables error-free long-distance transmission up to 10km, five times the distance achieved by traditional PSM optical modules.

200G QSFP-DD LR8

The 200G QSFP-DD LR8 optical module is based on an 8-channel LAN-WDM wavelength division multiplexing design, achieving a 1E-12 bit error rate over a transmission distance of 20km. In the context of high-reliability 5G middle-backhaul transmission systems, the long-term transmission of PAM4 signals may pose challenges such as insufficient bandwidth. The 200G QSFP-DD LR8, however, offers a high-performance, low-latency, and highly reliable channel solution.

GIGALIGHT has recently made significant strides in optical design, particularly achieving notable success in the development of 800G optical modules. The design includes the completion of 800G SR8/DR8. For more details, please refer to here. Moreover, GIGALIGHT has successfully mass-produced and deployed 8-channel optical products in niche markets globally, encompassing various fields such as 200G, 400G, and 800G. Visit our official website for further information.

29Dec, 2023

Global AI Server Market Surges to $50 Billion in 2023, Expected to Exceed 50% Share by 2027

Recently, market research organization IDC released its latest research report on the global server market. The report indicates that in 2022, the global server market size experienced a year-on-year growth of 20.1%, reaching $123.224 billion. It is anticipated that the market size will see a slight increase to $128.471 billion in 2023. The annual growth rates for the following four years are estimated to be 11.8%, 10.2%, 9.7%, and 8.9%, respectively. By 2027, the market size is projected to reach $189.139 billion.

Specifically, in 2022, global x86 server sales amounted to $110.955 billion, marking an 18.8% year-on-year increase. In contrast, non-x86 server sales surged by 32.8% year-on-year to $12.269 billion. In 2023, the growth in the global server market slowed down, with x86 server sales experiencing a mere 2.3% year-on-year increase to $110.955 billion. However, non-x86 server sales maintained a high year-on-year growth of 22.1%, reaching $14.984 billion. According to IDC’s projections, by 2027, x86 server sales will escalate to $165.568 billion, while non-x86 server sales will also increase to $23.571 billion.

Looking at the broad pattern, although non-x86 servers have sustained continuous rapid growth over the coming years, x86 servers remain the mainstream in the server market, maintaining a sales-based share of over 87.5%. Even by 2027, the market share of non-x86 servers will still be slightly under 12.5%, barely up from 11.66% in 2023, an increase of less than 1 percentage point.

However, significant shifts are occurring within the non-x86 server market segment. Over the past decade, around half of the non-x86 servers were comprised of IBM’s Power Systems and System z mainframes, with the rest being a combination of other proprietary servers and Arm servers. But with the rise of Arm servers among hyperscale vendors and cloud builders, the portion led by Arm in the non-x86 server segment is growing. Furthermore, with the emergence of RISC-V servers, this market segment is expected to undergo further changes.

Another dataset from IDC reveals that in 2020, IBM’s System z and Power revenues amounted to $4.98 billion. This suggests that in 2020, non-x86 server product sales, including Arm servers and other processor architectures, totaled $3.87 billion. Assuming certain scenarios for future IBM products, such as the upgrade cycles for Power10 and z16 machines in 2021 and 2022, and Power11 and z17 machines in 2025, it’s possible to anticipate a gradual decline in IBM server hardware sales from slightly below $5 billion in 2020 to $3.5 billion in 2026, potentially reaching $3.3 billion by 2027. Based on IDC’s data, other types of servers within the non-x86 segment, like Arm/RISC-V, are expected to grow at a robust pace, subject to investment cycles of hyperscale enterprises and cloud builders, maintaining a benchmark level. Inevitably, if adoption occurs, the annual sales of Arm and RISC-V servers (mostly Arm servers) could reach around $20 billion.

Moreover, due to numerous hyperscale enterprises and cloud builders focusing on customizing Arm server CPUs and AI coprocessors, there’s a wide range of choices available. This trend faces pressure not only from using X86 server CPUs and Nvidia GPUs for AI and other compute-intensive workloads but also from the widespread adoption of customized Arm servers.

Considering this, how AI server sales (primarily for training but also for inference) differ from sales of other types of servers?

IDC mentioned in its report: “In each subsequent quarter of 2022, inflation had a stronger direct impact on servers, with the average selling price (ASP) year-on-year growth rate rising to 29% in the second quarter of 2023, while server shipments mostly hovered around low double-digit year-on-year growth for most of 2022 but declined by 1.4% year-on-year in the fourth quarter of 2022 and by 10% in the first quarter of 2023, and now in the second quarter of 2023, it has again fallen by 19.9% year-on-year.”

Compared to the significant decrease in shipment volumes, the continuous rise in average selling price is primarily driven by very expensive AI training and inference nodes (each node costing hundreds of thousands of dollars with four or eight GPUs). AI servers and non-AI servers indeed need to be segregated as they represent very distinct segments in the market. Therefore, The Next Platform, based on IDC’s revenue forecast for servers from 2020 to 2027 (IDC data indicates the global AI server market size was $11.2 billion in 2021), attempted to classify them as follows:

The Next Platform believes that unless something slows down the growth of AI models or makes AI training and inference computations cheaper, it’s reasonable to expect that by 2026 or 2027, AI servers will account for approximately half of the entire server market revenue.

The model assumes a steep decline in non-AI server revenue by 11.2% starting in 2022, followed by moderate growth similar to GDP every year. By 2023, this situation is expected to improve, with only a 5% decline. Between 2022 and 2023, AI server revenue will experience an astonishing nearly fivefold leap, reaching around $50 billion in the global AI server market size by 2023. Subsequently, there will be a healthy and stable growth of around 20% in 2024, tapering off to 15% by 2027. According to this predictive model, AI server sales will surpass traditional server sales by 2027.

Currently, AI applications are experiencing explosive growth. With the increase in Nvidia GPU supply and price reductions, as well as the entry of GPUs from other brands and other types of AI accelerators gaining traction in the market, everything is expected to stabilize and normalize, possibly reaching an entirely new level.

One question remains: how much supply does the world’s AI need? Predicting the situation four or five years down the line is indeed very challenging. If the demand for AI accelerators continues to outpace supply, and prices remain high, revenue will stay elevated. If production doubles or triples and prices decrease by half or two-thirds, then revenue might stay unchanged.

Clearly, no manufacturer would want to lower profit margins through excessive sales, but intense competition might force companies into that position.

From 1985 to 2000, it took fifteen years for RISC/Unix machines and internet technology to capture a 45% share of server market revenue through active upgrades in mainframes and proprietary small-scale machines. It might take a similar fifteen-year span — from 2010 to 2025, or from 2011 to 2026 — for AI servers to account for around 45% of global server revenue, with AI workloads replacing or expanding into various applications you can imagine.

18Dec, 2023

High-Speed Optical Module Demand Soars: AI Computing and Market Projections Drive Innovations

Discovering the intersection of AI computing and escalating market trends, the reliance on optical modules has surged. From high-scale computational scenarios in AI-powered systems to market forecasts propelling technological advancements, the landscape is evolving. Let’s delve into the interplay of AI, market projections, and groundbreaking optical innovations, reshaping the future of computational networking

Section 1, AI Computing Scenario Optical Module Application

In the usage of NVIDIA’s A100 systems, such as within the NVIDIA DGX SuperPOD, concerning the application of optical modules required for computation and storage sides, all cables for both computation and storage sides utilize NVIDIA MF1S00-HxxxE series AOC active optical cables. Consequently, each port requires an optical module, meaning each cable corresponds to two optical modules. Therefore, the computation and storage sides total required amounts to (1120+1124+1120)*2 + (280+92+288)*2 = 8048 optical modules. In other words, the approximate quantity of 200G optical modules needed per GPU is about 1:7.2.

DGX GH200 supercomputer, equipped with 256 NVIDIA GH200 Grace Hopper super chips, each of which can be regarded as a server, interconnected through NVLINK SWITCH. Structurally, the DGX GH200 adopts a two-tiered fat tree topology, with the first and second tiers employing 96 and 36 switches, respectively. Each NVLINK SWITCH has 32 ports operating at a speed of 800G. Additionally, the DGX GH200 is equipped with 24 NVIDIA Quantum-2 QM9700 IB switches for the IB network.

For estimation based on ports, given the closer proximity of L1 tier, it’s assumed that copper cables are used for connectivity without involving optical modules. For the L2 tier with 36 switches in a non-converging fat tree architecture, each switch’s ports connect downwards to the uplink ports of L1 tier switches. Therefore, a total of 36 * 32 * 2 = 1152 units of 800G optical modules are needed. In the IB network architecture, the 24 switches require 24 * 32 = 768 units of 800G optical modules. Thus, the DGX GH200 supercomputer requires a total of 1152 + 768 = 1920 units of 800G optical modules, corresponding to 12 units of 800G optical modules per GH200 chip.

Section 2, Optical Module Market Demand Forecast Driven by Computational Networking

The significant surge in computing power demand driven by new-generation information technologies like cloud computing, artificial intelligence, and big data has hastened the development of cloud computing infrastructure. Concurrently, with the comprehensive launch of projects such as the “Eastern Data and Western Computing,” China’s computational infrastructure construction is rapidly advancing. Optical modules, as pivotal components in cloud computing data centers, will experience a continuous surge in market demand due to the substantial increase in data transmission volume.

According to LightCounting’s projections, the global adoption rate of 800G technology is expected to be merely 0.62% by 2023. Large AI models like ChatGPT impose new demands on data flows within and outside data centers, potentially accelerating the transition of optical modules towards 800G. It’s anticipated that 800G optical modules will start dominating the market by the end of 2025.

Based on LightCounting’s data, the global market size for optical modules increased from $58.6 billion in 2016 to $66.7 billion in 2020. Projections suggest that by 2025, the global optical module market will reach $11.3 billion, 1.7 times the size in 2020. Structurally, the data communication market is expected to hold around 60%, while the telecommunications market will encompass approximately 40% of the market share.

GIGALIGHT Optical Modules for Computational Applications

The core technology of GIGALIGHT’s silicon optical modules lies in its innovative packaging design with high-coupling efficiency in the free-space COB and the MZI software locking algorithm. Additionally, in terms of silicon waveguide cores, the company has engaged in collaborative design work for multiple silicon chip designs with partners.