AI Cluster Networking Report 2025 Released
95% of operators say simulating real-world workloads is crucial.
The pace of AI adoption appears to be faster than infrastructure expansion, highlighting the need for a shift from simple AI expansion to an "optimization" strategy.
Keysight Technologies, in collaboration with market research firm Heavy Reading, released the "AI Cluster Networking Report 2025 (Beyond the Bottleneck)", which found that AI adoption is outpacing infrastructure expansion. The report emphasized that telecommunications and cloud service providers must shift their strategies from simple infrastructure expansion to one focused on optimization.
As AI adoption accelerates across industries, demand for data center infrastructure is skyrocketing. However, 62% of respondents stated that they aim to extract more value from existing infrastructure without incurring new capital expenditures, demonstrating the rise of efficiency-focused operational strategies.
Specifically, 95% of respondents said strategies to emulate real-world AI workloads are important, with a significant number citing the lack of realistic-scale simulation tools as a key challenge. To meet this demand, Keysight is providing solutions that support emulation and verification of large-scale AI infrastructure.
According to the report, investment in AI infrastructure is expected to continue. Eighty-nine percent of respondents plan to maintain or increase their investments next year, with cloud integration (51%), accelerated GPU adoption (49%), and high-speed network upgrades (45%) cited as key growth drivers. In fact, 55% of operators have deployed 400G interconnects, and adoption of 800G and 1.6T technologies is also gradually expanding.
Interest in high-performance networking technologies is also growing. Fifty-eight percent of respondents are evaluating Ultra Ethernet, and network bottlenecks are emerging as a key challenge for AI expansion. Consequently, securing network capacity is emerging as a key factor directly related to future AI scalability.
“AI data centers are no longer about mere performance or scale,” said Ram Periyakarupan, vice president of Keysight’s Network Applications & Security Group. “Operators need deeper insights, sophisticated validation, and smarter infrastructure choices.”
This survey was conducted from March to April 2025 among professionals from major telecommunications and cloud service providers in North America, Asia Pacific, and Europe.