AIDC’s Next Battleground: Networks
Why AI Data Center Growth Is Shifting Toward Connectivity and Optical Communications
The AI infrastructure investment cycle is entering a new phase.
Until recently, the market’s attention was concentrated on power infrastructure, including electricity generation, substations, transformers, and distribution equipment. However, as AI data centers (AIDCs) expand, the next bottleneck is increasingly becoming network connectivity.
AI Data Centers → Distributed Computing → More Connectivity → Higher Bandwidth → Optical Communications
1. The AI Data Center Investment Theme Is Expanding
The first stage of the AIDC investment cycle focused heavily on securing electricity.
Large-scale AI data centers require enormous amounts of power, and the long lead times for substations, transmission facilities, and transformers created a major supply bottleneck.
This led power equipment companies to become some of the earliest beneficiaries of the AI infrastructure boom.
As of August 31, 2026, LS Electric, Hyosung Heavy Industries, and HD Hyundai Electric had recorded significant share-price gains, with their average 2026E P/E reaching approximately 40.9x.
By comparison, Korea's three major telecom operators were trading at an average 2026E P/E of approximately 11.0x.
This valuation gap is important.
The report argues that investors should move beyond simply asking:
"Which industry is growing fastest?"
Instead, investors should consider:
"Which part of the AIDC value chain has strong future growth that has not yet been fully reflected in valuations?"
That is where networks become important.
2. AI Training and AI Inference Require Different Data Center Locations
One of the most important structural changes is the geographical separation between AI training and AI inference.
AI Training
Training large AI models requires enormous computing resources and power, but it is relatively insensitive to latency.
Therefore, training-oriented AIDCs can be located in regions outside the Seoul metropolitan area where electricity is more readily available.
AI Inference
Inference is fundamentally different.
Services such as:
- Robotics
- Autonomous driving
- Conversational AI
- Real-time edge AI
- Physical AI
must respond quickly to user or physical-world inputs.
Consequently, inference workloads need to be located much closer to users.
The report estimates that AI inference could increase its share of total AI demand from 25% in 2025 to 43% in 2030, surpassing training demand from 2027.
This creates an important infrastructure problem.
Training may be located outside metropolitan areas, while inference remains concentrated near major population centers.
These facilities therefore need to be connected.
3. Connectivity Becomes the Next AIDC Bottleneck
As AI workloads become geographically distributed, the traditional data center network architecture is changing.
Traditional data centers primarily handled north-south traffic, meaning communication between external users and servers.
AIDC networks increasingly require east-west traffic between:
- GPUs
- GPU servers
- Racks
- Clusters
- Data centers
The report describes this as a fundamental change in network architecture.
This means the importance of the following infrastructure will increase:
- Data Center Interconnects (DCI)
- Backbone networks
- Dedicated lines
- DWDM optical transport
- Submarine cables
- Optical transceivers
- High-speed switches
Korea's submarine cable demand connecting the country is projected to increase from 92Tbps in 2026 to 295Tbps in 2031.
AI-related demand alone is expected to increase from 24Tbps to 198Tbps over the same period.
That represents roughly an eightfold increase in AI-related demand.
4. 400G → 800G → 1.6T: Optical Communication Is Entering a New Cycle
AI computing is moving from individual GPUs toward increasingly interconnected GPU clusters.
As the number of GPUs increases, the number of connections also increases.
At the same time, bandwidth requirements per connection continue to rise.
The report expects networking speeds to progress from:
400G → 800G → 1.6T
This should create structural demand for high-speed optical transceivers and optical fiber.
The important point is that AI infrastructure does not simply require more servers.
It requires:
More GPUs × More connections × Higher bandwidth
The report identifies this combination as a key driver of the telecom equipment market.
5. Why Optical Transceivers Are Becoming More Important
As transmission speeds increase, signal integrity becomes more difficult to maintain.
The report explains that higher-speed transmission reduces the available link budget, meaning the maximum amount of signal loss that can be tolerated while maintaining reliable communication.
At higher data rates, even relatively small distortions can create errors.
As a result, networks increasingly require shorter transmission links and more optical regeneration points.
This can increase the number of optical transceivers required throughout the network.
This creates a potentially powerful structural demand driver for optical communications companies.
6. CPO Could Become the Long-Term Next Step
Another major technology highlighted in the report is Co-Packaged Optics (CPO).
Traditional pluggable optical transceivers have a relatively long electrical signal path between the switch ASIC and optical module.
According to the report, this path can create approximately 22dB of loss, requiring DSP technology to compensate for the loss.
CPO places the optical engine much closer to the switch ASIC, substantially shortening the electrical signal path.
The report estimates that this could reduce losses to approximately 4dB.
This is particularly important as AI networks move toward higher transmission speeds.
The report does not view CPO as simply replacing existing optical communications demand.
Instead, CPO changes the location and architecture of optical connectivity within the system, while fiber connections between racks and clusters remain necessary.
7. The Three Korean Telecom Operators Are Strategically Important
The report highlights SK Telecom, KT, and LG U+ as the three major Korean telecom operators positioned to benefit from the AIDC expansion.
Their advantage is not simply the amount of data center capacity they own.
They also control or have access to:
- Metropolitan data centers
- Backbone networks
- Enterprise networks
- Submarine cables
- Edge infrastructure
- Existing telecom facilities
This combination allows them to provide both data center capacity and connectivity.
The three operators currently hold an estimated combined AIDC capacity of approximately 550MW, representing about 29% of Korea's total domestic data center receiving capacity.
8. SK Telecom: The Most Aggressive AIDC Expansion Strategy
SK Telecom has established particularly ambitious AIDC targets.
Its strategy includes:
- 5GW of AIDC capacity by 2035
- 15GW in the longer term
- GW-scale data centers in non-metropolitan regions
- Expansion through SK Hyper
- Establishment of SK Horizon
- External funding and project financing
SK Telecom is also pursuing a vertically integrated strategy involving land, power, EPC, and cooling infrastructure.
The report highlights this vertical integration as one of SKT's key differentiating factors.
The company is particularly focused on large-scale AIDCs outside the Seoul metropolitan area, where power availability is more favorable.
This strategy is closely aligned with the expected growth of AI training workloads.
9. KT: A Balanced Metropolitan and Non-Metropolitan Strategy
KT is targeting approximately 1GW of additional AIDC capacity by 2031.
Rather than owning the entire capacity, KT plans to combine:
- Direct ownership
- Leasing
- Partnerships
- External investment
KT is also planning to leverage approximately 3,500 telecom facilities nationwide for AI edge infrastructure.
Its strategy divides workloads geographically:
Metropolitan areas → AI inference
Non-metropolitan areas → AI training
This makes KT's strategy relatively flexible because it can align infrastructure deployment with the actual demand characteristics of different AI workloads.
10. LG U+: An Asset-Light AIDC Strategy
LG U+ is targeting approximately 400MW of AIDC capacity by 2030.
Its strategy is more focused on the Seoul metropolitan area, particularly:
- Paju
- Pyeongchon
- Gasan
LG U+ is also expanding its cooperation with asset management companies.
The report highlights its partnership-based DBO (Design, Build, Operate) approach as a way to secure operating revenue while reducing the burden of directly funding all infrastructure investments.
This gives LG U+ a different investment profile from SKT's aggressive development strategy.
11. Why Seoul Metropolitan AIDCs Could Become More Valuable
The shortage of suitable power infrastructure is making new metropolitan AIDC construction increasingly difficult.
According to the report, as of March 2026:
- 522 applications had entered the preliminary technical review process.
- 243 passed the first phase.
- 24 proceeded to the main review.
- Only 10 received final supply approval.
This resulted in a final approval rate of approximately 1.9% of total applications.
This has an important implication.
If new metropolitan AIDC supply becomes difficult to create while inference demand continues increasing, the value of existing high-quality data center capacity can rise.
The report therefore expects metropolitan AIDC usage fees to increase structurally as contracts are renewed.
12. AIDC Investment Is Moving From Power to Networks
The overall investment cycle can now be viewed in several stages.
| AIDC Investment Stage | Key Infrastructure |
| Stage 1 | Power generation |
| Stage 2 | Transformers & substations |
| Stage 3 | Data center construction |
| Stage 4 | DCI & backbone networks |
| Stage 5 | 400G / 800G optical communication |
| Stage 6 | 1.6T optical interconnects |
| Stage 7 | CPO and advanced optical architecture |
This does not mean the power infrastructure theme is finished.
Instead, the investment opportunity is expanding into the network layer.
The report specifically argues that investors should focus on the gap between expected growth and what is already reflected in stock valuations.
13. Key Investment Candidates
The report's company rating table gives:
| Company | Rating | Target Price |
| SK Telecom | BUY | ₩118,000 |
| KT | BUY | ₩73,000 |
| LG U+ | BUY | ₩21,000 |
| Taihan Fiberoptics | Not Rated | — |
| OE Solutions | Not Rated | — |
These ratings and target prices are from the September 9, 2026 Yuanta Securities report and should not be interpreted as a current independent investment recommendation.
The report's broader investment framework is particularly interesting because it does not simply focus on telecom operators.
The expansion of AIDC connectivity could also benefit companies involved in:
- Optical transceivers
- Optical fiber
- DCI equipment
- AI switches
- NICs
- DWDM
- CPO
- Submarine cable infrastructure
14. The Bigger Picture: AI Performance Depends on Networking
One of the most important conclusions from the report is that AI infrastructure performance is no longer determined by computing power alone.
As GPU clusters become larger, the network connecting those GPUs becomes increasingly important.
If data cannot move quickly enough:
GPU utilization falls → latency rises → computing resources are wasted → AI service costs increase.
This is particularly important for AI inference.
The report notes that KV-cache transfers between GPUs can become a significant bandwidth bottleneck, creating demand for 400G, 800G and ultimately 1.6T optical interconnects.
In other words:
The next bottleneck in AI infrastructure may not be computing power. It may be the ability to move data between computing resources fast enough.
15. Final Investment Takeaway
The AIDC investment story is entering a new phase.
The first wave was dominated by power infrastructure.
The next wave could increasingly focus on connectivity and network equipment.
The structural chain is clear:
AI model expansion
↓
More GPUs
↓
More GPU-to-GPU communication
↓
More east-west traffic
↓
Higher DCI demand
↓
400G → 800G → 1.6T
↓
Higher optical transceiver demand
↓
CPO adoption
↓
Expansion of optical communications infrastructure
Korea is particularly interesting because AI training facilities are likely to expand toward regions with available power, while inference infrastructure needs to remain close to metropolitan users.
That geographic separation creates a natural requirement for high-capacity, low-latency networks connecting
multiple AI hubs.
Therefore, the AIDC investment theme should no longer be viewed simply as a power infrastructure story.
It is increasingly becoming a network infrastructure story.
And within that network story, DCI, optical transceivers, optical fiber, submarine cables, 800G/1.6T connectivity, and CPO are emerging as some of the most important technologies to watch.
Source: Yuanta Securities, “AIDC’s Next Battleground: Networks,” Telecom Services Sector Report
