Subscribe to the Data Center Knowledge Newsletter
Get analysis and expert insight on the latest in data center business and technology delivered to your inbox daily.
Data Center Networking Needs Are Changing Thanks to AI – ReportData Center Networking Needs Are Changing Thanks to AI – Report
Soaring AI workloads are transforming modern data center networks, a new survey reveals.
4 Min Read
Data center networks are being optimized to support AI-driven demands and evolving workloads.Image: Alamy
AI is having an enormous impact on data centers – from construction and design to power and compute. Now, data center networking needs are also evolving to keep pace.
The energy impact of AI on data centers has been well documented, with power requirements playing a significant role in sustainability. AI is also accelerating demand for new data center buildouts. Another core area where AI is impacting data centers is the network.
A new global survey commissioned by Ciena Corporation reveals that AI is reshaping data center networking infrastructure at an unprecedented scale. The study, which gathered insights from over 1,300 data center decision-makers across 13 countries, highlights how AI’s growing prominence is reshaping data center network architecture planning and implementation.
AI Networking Survey: Key Findings
According to the Ciena survey, 43% of new data center facilities are expected to be dedicated to AI workloads. In addition:
Data center experts predict at least a 6x increase in Data Center Interconnect (DCI) bandwidth demand over the next five years
53% of respondents believe AI workloads will place the biggest demand on DCI infrastructure in the next two to three years
87% of participants anticipate needing 800 Gb/s or higher per wavelength for fiber-optic capacity
98% view pluggable optics as important for reducing power consumption and physical footprint
“It was always clear to us that AI was going to drive a transformation in DCI infrastructure. The question was really to what extent,” Jürgen Hatheier, international chief technology officer at Ciena, told Data Center Knowledge.
“Typically, annual growth of broadband network traffic has been in the region of 20-30%, but our survey found that the expectation is for this growth rate to potentially double because of AI applications.”
Ciena isn’t the only group reporting a large spike in networking demand related to AI either. Analyst firm Dell’Oro has reported that there has been record-breaking data with sales, fueled largely by AI demand.
“In 2024, more than 90% of the year-on-year increase in data center switch sales was attributable to AI buildout,” said Sameh Boujelbene, VP of data center switch and AI networks research at Dell’Oro.
“We expect this trend to continue as Ethernet gains momentum in AI networks.”
Sustainability Concerns Drive Interest in Pluggable Optics
As bandwidth requirements skyrocket, data center operators are increasingly focused on sustainable approaches to network expansion.
The Ciena survey found near-universal (98%) agreement among data center experts that pluggable optics represent an important technology for reducing power consumption and network infrastructure’s physical footprint. Pluggable optics are modular devices for optical data transport.
“The benefits of pluggable optics are about flexibility in scaling network designs and, of course, power efficiencies,” Hatheier said. “As capacity scales to higher rates, traditional data center technologies will start to hit physical limits, and coherent technology will make its way into and around the data center, reducing hardware footprint and the overall power consumption of the equipment.”
The CTO noted that what the power and space savings will look like in practice will depend on the use cases. He added that the vast majority of power draw for data centers, particularly those dedicated to AI, will be from inferencing distributed across networks and optics, at the very least, bringing greater efficiency to handle that traffic.
Beyond Bandwidth: What Else Do Data Center Networks Need for AI?
While more bandwidth is needed for the next generation of so-called ‘AI factories,’ it’s only part of the equation.
Hatheier noted that AI comes with new network requirements, diverse traffic types, and dynamic traffic patterns that can’t be handled by simply adding more hardware to increase network capacity.
In his view, smart networks that dynamically adapt to specific demands at any given time are needed.
“Intelligent automation platforms… can drive closed-loop optimization, ensuring AI traffic is prioritized and routed efficiently without manual intervention," Hatheier said. “Multi-layer automation across optical and IP layers can dynamically adjust bandwidth, optimize power consumption and prevent congestion in real time.”
Read more of the latest data center networking news
Hatheier also noted that network slicing will complement automated frameworks by allowing operators to create customized virtual networks tailored to the specific requirements of AI applications. Each slice can be optimized for critical parameters such as latency, throughput, and security, ensuring dedicated resources for high-priority AI tasks.
“Combined with automation, network slicing enables data centers to dynamically allocate, reconfigure, and manage resources on demand, providing a scalable, adaptive, and cost-efficient environment that fully supports the evolving needs of AI workloads,” he said.
About the Author
You May Also Like
