AI is having an unlimited impression on information facilities – from building and design to energy and compute. Now, information middle networking wants are additionally evolving to maintain tempo.
The vitality impression of AI on information facilities has been properly documented, with energy necessities taking part in a big function in sustainability. AI can also be accelerating demand for brand spanking new information middle buildouts. One other core space the place AI is impacting information facilities is the community.
A brand new international survey commissioned by Ciena Company reveals that AI is reshaping information middle networking infrastructure at an unprecedented scale. The examine, which gathered insights from over 1,300 information middle decision-makers throughout 13 international locations, highlights how AI’s rising prominence is reshaping information middle community structure planning and implementation.
AI Networking Survey: Key Findings
Based on the Ciena survey, 43% of recent information middle amenities are anticipated to be devoted to AI workloads. As well as:
-
Information middle specialists predict at the least a 6x enhance in Information Middle Interconnect (DCI) bandwidth demand over the subsequent 5 years
-
53% of respondents imagine AI workloads will place the largest demand on DCI infrastructure within the subsequent two to 3 years
-
87% of individuals anticipate needing 800 Gb/s or greater per wavelength for fiber-optic capability
-
98% view pluggable optics as essential for decreasing energy consumption and bodily footprint
“It was all the time clear to us that AI was going to drive a metamorphosis in DCI infrastructure. The query was actually to what extent,” Jürgen Hatheier, chief expertise officer at Ciena, informed DCN.
“Usually, annual development of broadband community visitors has been within the area of 20-30%, however our survey discovered that the expectation is for this development fee to doubtlessly double due to AI purposes.”
Ciena isn’t the one group reporting a big spike in networking demand associated to AI both. Analyst agency Dell’Oro has reported that there was record-breaking information with gross sales, fueled largely by AI demand.
“In 2024, greater than 90% of the year-on-year enhance in information middle swap gross sales was attributable to AI buildout,” stated Sameh Boujelbene, VP of knowledge middle swap and AI networks analysis at Dell’Oro.
“We count on this pattern to proceed as Ethernet good points momentum in AI networks.”
Sustainability Issues Drive Curiosity in Pluggable Optics
As bandwidth necessities skyrocket, information middle operators are more and more centered on sustainable approaches to community growth.
The Ciena survey discovered near-universal (98%) settlement amongst information middle specialists that pluggable optics symbolize an essential expertise for decreasing energy consumption and community infrastructure’s bodily footprint. Pluggable optics are modular units for optical information transport.
“The advantages of pluggable optics are about flexibility in scaling community designs and, in fact, energy efficiencies,” Hatheier stated. “As capability scales to greater charges, conventional information middle applied sciences will begin to hit bodily limits, and coherent expertise will make its means into and across the information middle, decreasing {hardware} footprint and the general energy consumption of the tools.”
The CTO famous that what the ability and house financial savings will seem like in observe will rely on the use instances. He added that the overwhelming majority of energy draw for information facilities, significantly these devoted to AI, will probably be from inferencing distributed throughout networks and optics, on the very least, bringing better effectivity to deal with that visitors.
Past Bandwidth: What Else Do Information Middle Networks Want for AI?
Whereas extra bandwidth is required for the subsequent era of so-called ‘AI factories,’ it’s solely a part of the equation.
Hatheier famous that AI comes with new community necessities, various visitors varieties, and dynamic visitors patterns that may’t be dealt with by merely including extra {hardware} to extend community capability.
In his view, good networks that dynamically adapt to particular calls for at any given time are wanted.
“Clever automation platforms… can drive closed-loop optimization, making certain AI visitors is prioritized and routed effectively with out handbook intervention,” Hatheier stated. “Multi-layer automation throughout optical and IP layers can dynamically regulate bandwidth, optimize energy consumption and forestall congestion in actual time.”
Learn extra of the most recent information middle networking information
Hatheier additionally famous that community slicing will complement automated frameworks by permitting operators to create personalized digital networks tailor-made to the particular necessities of AI purposes. Every slice could be optimized for essential parameters comparable to latency, throughput, and safety, making certain devoted assets for high-priority AI duties.
“Mixed with automation, community slicing permits information facilities to dynamically allocate, reconfigure, and handle assets on demand, offering a scalable, adaptive, and cost-efficient atmosphere that totally helps the evolving wants of AI workloads,” he stated.
