What is the purpose of a optical transceiver?
Dec 23, 2025|
Many people may be unfamiliar with the term "optical transceiver." But every time you browse TikTok, make video calls, or store files in cloud storage, this little device is silently working behind the scenes.
I've worked in the telecommunications industry for over ten years, from initially working with 1G SFP modules to now traveling the world promoting 400G and 800G solutions, witnessing firsthand the industry's explosive growth. Today, I'll talk about what optical transceivers are actually used for, trying to explain it in simple terms and avoid jargon.
First, let's understand what an optical transceiver is
An optical transceiver, generally called an "optical module" in the industry, is essentially a signal converter.
Your home router, company switch, and data center servers all run on electrical signals. But electrical signals have a major problem-they don't travel very far. With copper cables, a 10G signal can only travel 30-50 meters before it breaks down, and it's extremely susceptible to interference; even a nearby motor running can cause problems.
Optical signals are different. Theoretically, optical signals in single-mode fiber can travel hundreds of kilometers, with an absurdly large bandwidth; a single fiber as thin as a human hair can transmit tens of terabytes of data simultaneously.
But here's the problem: devices can't process optical signals.
That's why we need optical modules as "translators":
When sending: converting electrical signals into optical signals and sending them into the fiber.
When receiving: converting the optical signals in the fiber back into electrical signals for the devices to process-it's that simple.
Data Centers – The Largest Consumers of Optical Modules
Let me start with some concrete figures. Last year, I had the opportunity to visit a leading cloud provider's data center in Zhangbei. The operations and maintenance staff told me that their single campus had over 500,000 optical modules, and they replaced several hundred every month for various reasons; spare parts were piled up like small mountains in the warehouse.
And this was just one campus. The combined total of several major domestic manufacturers easily exceeds ten million optical modules in operation.
Server Access
In mainstream data centers today, each server is equipped with at least two 25G or 100G network ports, all of which require optical modules. A rack housing 40 servers would require 80 optical modules just for that server layer.
Some people ask, "Why not just use copper cables for short distances?"
There is indeed such a solution, called DAC (Direct Connect Copper Cable), which is indeed cheap and effective within 3 meters. However, it doesn't work beyond 3 meters due to severe signal attenuation. Data center cabling is rarely neat and tidy; it often involves twists and turns, with 5 or 10 meters being quite common. In such cases, optical modules are essential.
Spine-Leaf Interconnect Architecture
Most decent data centers now use a Spine-Leaf architecture. Leaf switches manage server access, while Spine switches handle east-west traffic forwarding.
The distance between the Leaf and Spine varies from tens to a couple of hundred meters, and is generally 100G or higher, with major manufacturers already moving to 400G.
According to data from LightCounting in early 2024, 100G optical modules are still the largest category in data center shipments, but 400G is experiencing astonishing growth, increasing by nearly 80% year-on-year.
I feel that by 2025, 400G will become the standard for newly built data centers.
Data Center Interconnect (DCI)
Large companies typically have several data centers in a city, requiring high-speed interconnection for data synchronization and disaster recovery.
DCI distances within the same city are usually 10-80 kilometers. Previously, 100G LR4 and ER4 were used for this scenario, but now 400G ZR is increasingly being adopted. ZR is a coherent optical module, capable of operating at distances of 80 kilometers or even longer, with a single-wavelength 400G, which is very powerful.
Last year, a client wanted to establish a 400G direct connection between two data centers 60 kilometers apart. Initially, the plan was to use traditional DWDM equipment, which would have cost several million yuan. Later, they switched to a direct connection with 400G ZR optical modules, cutting the cost by more than half and simplifying maintenance considerably. This is a testament to the benefits of technological advancement.
AI Clusters – The Hottest Trend Recently
Large-scale models have become popular in the last two years, and the network bandwidth requirements for training clusters are simply insane.
NVIDIA's DGX H100 server, with 8 GPUs per machine and each GPU equipped with a 400G Ethernet port, requires eight 400G optical modules per machine. Setting up a cluster of tens of thousands of GPUs would result in astronomical costs for optical modules.
It's rumored that a major manufacturer prepaid hundreds of millions to its suppliers to secure 800G optical module production capacity.
Personally, I feel that the demand for AI has come too fast, and the optical module supply chain has been consistently tight. The most direct evidence is that the stock prices of several leading optical module manufacturers have skyrocketed this year.
Telecommunications operator networks are a traditional market for optical modules. Although not as "sexy" as data centers, their scale is stable.
5G Transport Network
5G base stations are divided into three levels: AAU, DU, and CU. The connections between them are called fronthaul, midhaul, and backhaul.
Fronthaul (AAU to DU) most commonly uses 25G optical modules, with distances generally not exceeding 20 kilometers. This segment has extremely high requirements for latency and synchronization, using the eCPRI protocol, and the optical modules also have some special requirements. Last year, in a 5G fronthaul project with a provincial mobile operator, they were very strict with latency testing of the optical modules; several batches were returned due to excessive latency. Quality control is crucial in telecom projects.
Midhaul and backhaul use optical modules with higher speeds, including 50G and 100G, and over much longer distances, potentially tens of kilometers.
The peak of 5G deployment has actually passed, but 6G is in pre-research, so there are still opportunities later.
Metropolitan Area Networks (MANs) and Backbone Networks
Metropolitan area networks (MANs) are primarily the networks operated by carriers within cities, aggregating various access traffic and sending it up to the backbone network.
The backbone network is a long-distance transmission network spanning cities and provinces, carrying almost all internet traffic. Backbone networks must use DWDM systems, cramming dozens or even hundreds of wavelengths into a single optical fiber, each wavelength running at 100G or 400G.
The optical modules used in this area are the most technologically advanced, primarily coherent optical modules, and are expensive; a single module can easily cost tens of thousands of yuan. Frankly, the backbone network business has high profit margins, but the volume is small, and the customer base is limited to a few carriers, making relationships crucial.
Enterprise network optical module requirements
Slightly larger companies will definitely need to run fiber optic cables between office buildings. The most extreme example I've seen is the network of a car factory campus. The factory area is so large that some buildings are three or four kilometers apart, requiring 10G LR or even ER optical modules.
Enterprise customers are generally price-sensitive. Original equipment manufacturer (OEM) optical modules are too expensive, so most will choose third-party compatible modules. As long as you find a reliable supplier, compatible modules generally work without problems. However, there are exceptions. Some large state-owned enterprises and financial institutions require the use of OEM modules in their procurement processes, even if they are two or three times more expensive. Compliance requirements are there; there's no way around it.
In enterprise data centers, optical modules are also needed to interconnect servers and storage devices.
There are two main systems for storage networks: Fibre Channel (FC) and Ethernet. FC is an older protocol, but it's still widely used in industries like finance and healthcare due to its stability and reliability.
FC optical modules have their own specifications: 8G, 16G, and 32G FC, and they cannot be used interchangeably with Ethernet optical modules. In recent years, the NVMe-oF protocol has gained popularity, using Ethernet to carry storage traffic, and FC's market share is gradually being eroded. However, this process will be very slow because the existing market is too large.
Other Applications of Optical Modules
Factory environments are harsh, with high levels of electromagnetic interference and drastic temperature fluctuations, which ordinary optical modules cannot withstand. Industrial-grade optical modules require an operating temperature range of -40°C to +85°C, as well as vibration and shock resistance. The cost is significantly higher than ordinary optical modules, but industrial customers are not concerned about the extra cost; they prioritize stability.
A friend who works on a steel mill project told me that ordinary switches are simply not usable for the network equipment near their blast furnaces; they have to use industrial-grade equipment with industrial-grade optical modules, otherwise the network will overheat and crash.
TV stations also use optical modules to transmit video signals internally, but the protocol is slightly different; it's SDI over Fiber.
4K and 8K ultra-high-definition signals have very large bandwidths, and compression introduces latency, which is unacceptable for live broadcasts. Therefore, the broadcasting industry uses uncompressed transmission, which places very high demands on the bandwidth of optical modules.
Military optical modules are a completely different world, requiring various hardening and certifications, and the price is also in a completely different league-outrageously expensive.Specific details are not convenient to disclose, but in short, the technical barriers are very high, and there are not many players who can enter this field.
How to Choose an Optical Module?
Having discussed so many uses, how do you choose an optical module in actual work?
SR: Within 100 meters, use multimode fiber
DR: 500 meters, single-mode fiber
FR: 2 kilometers, single-mode fiber
LR: 10 kilometers, single-mode fiber
ER: 40 kilometers
ZR: 80 kilometers or even longer.
Leave some margin when selecting optical modules. For example, if the measured distance is 800 meters, choosing
DR (500-meter specification) is definitely not enough; you need to use FR.
Multimode optical modules can only be used with multimode fiber, and single-mode optical modules can only be used with single-mode fiber. If you choose the wrong type, the module will not light up.
Multimode fiber has several grades: OM1, OM2, OM3, OM4, and OM5. The higher the grade, the longer the supported distance. Currently, OM3 and OM4 are the mainstream. Single-mode fiber is basically G.652, so you don't need to worry about the model.
Although optical modules have the MSA standard, different equipment manufacturers still implement various methods, so not all are necessarily compatible. Cisco and Huawei equipment have more restrictions on third-party optical modules, and some require command-line input for recognition. Arista and Mellanox are relatively more open. To be on the safe side, ask the supplier if they have tested it on the target device. Major compatible optical module manufacturers usually have compatibility lists.
High-speed optical modules are consuming increasingly more power. A 400G DR4 module consumes 8-10W, while a 400G ZR module can reach 15-20W.
If all the optical modules are installed in a switch, the total power consumption can be several hundred watts, posing a challenge to heat dissipation. Remember to factor this into your design, to avoid overloading the data center's cooling system.
800G modules are currently in high demand, with some models having delivery times of three to four months. If a project has a tight schedule, it's essential to secure supplies in advance.
Troubleshooting Optical Modules
Start with the simplest: Is the optical module plugged in securely? Is the fiber optic cable connected correctly? Don't laugh, some people actually don't hear a "click" when plugging in the fiber optic cable and think it's connected correctly when it's not. Then check the fiber polarity. For dual-fiber connections, the transmitter (Tx) should be connected to the receiver (Rx); connecting them reversed will prevent it from lighting up. If it still doesn't work, use an optical power meter to measure the transmit and receive power to see if one of the optical modules is faulty.
This situation is more complicated and could have many causes:
Insufficient received optical power (high fiber loss, dirty connectors)
Excessive bending of the fiber (especially single-mode fiber; a too-small bending radius will cause light leakage)
A problem with the optical module itself
A problem with the port
Trace the link to see which segment of the fiber is faulty. If you still can't find the problem, try replacing the optical module, the fiber, or the port-using the process of elimination.
Let's talk about some technology trends
800G and 1.6T:
400G is currently mainstream, while 800G is already seeing mass production. In 2024, the shipment volume of 800G optical modules reached two to three million units.
1.6T is also under development, and small-batch shipments began in 2025. The speed improvement is ridiculously rapid.
Silicon photonics has been hyped for many years.
Frankly, I personally think it's been over-marketed. Theoretically, silicon photonics can reduce costs and increase integration, but in actual mass production, yield issues have not been completely resolved. Furthermore, silicon-based materials cannot be used to make lasers; they still need to be mixed and integrated with III-V group materials.
Of course, this is just my opinion; many in the industry disagree. Intel and Cisco are heavily promoting silicon photonics, and they must have their reasons.
This concept is more radical: it directly packages the optical engine and the switching chip together. The advantage is a significant reduction in power consumption and an increase in bandwidth density. The disadvantage is that the optical module cannot be replaced individually; if one fails, the entire board may need to be replaced.
Major companies like Google and Meta are heavily promoting CPO, and actual deployments are expected in 2025 or 2026. However, whether it will become mainstream is still uncertain. Maintenance colleagues dread CPO: how do you replace it if it fails? Should the entire system be decommissioned?
At Last
The core idea is simple: optical transceivers are the cornerstone of modern communication networks, with extremely wide applications.
From your home broadband ONU to the backbone network of telecom operators; from enterprise office networks to hyperscale data centers; from 5G base stations to AI training clusters-optical modules are everywhere.
This industry isn't glamorous, and the technical barriers aren't as intimidating as those in the chip industry, but its strength lies in its steady, continuous growth. The AI wave has given the industry a significant boost. If you're a network engineer, data center operations engineer, or simply interested in the communications industry, spending some time learning about optical modules is definitely worthwhile. The longer you work in this field, the more you'll appreciate this.