AI cluster networking: Paving the way for a transformational 2025

Keysight Technologies, Inc. and Heavy Reading have shared a pivotal 2025 report on AI cluster networking. As artificial intelligence adoption outpaces infrastructure development, telecom and cloud providers are urged to pivot from expansion to optimisation to handle next-generation AI tasks.

AI growth in various industries increases demands on data centres. However, traditional expansion initiatives seem inadequate. A significant 62% of respondents prefer maximising current infrastructure over new investments. This prompts operators to embrace performance optimisation strategies, such as real-world AI workload emulation to validate and enhance deployment efficiency for AI clusters.

The report, which drew insights primarily from industry respondents, showed 89% planning to either expand or maintain AI infrastructure investments. The predominant factors propelling this trend include cloud integration (on the rise at 51%), faster GPUs' deployment (49%), and high-speed network upgrades (45%).

Important findings from the report, titled Beyond the Bottleneck: AI Cluster Networking Report 2025, include

• Optimisation First Approach: Investment persists, but 62% say they focus on extracting value from current infrastructure sans new capital expenditures.
• Emulation Becomes Essential: A steep 95% emphasise the need for real-world workload emulation, despite lacking requisite simulation tools.
• Rising Infrastructure Pressure: Budget constraints (59%), infrastructure limitations (55%), and talent shortages (51%) are major hurdles.
• High-Speed Networking Expansion: Technologies like 800G, 1.6T, and Ultra Ethernet are explored or evaluated, reflecting growing momentum.
• Network Bottlenecks at the Forefront: An increasing interest in 1.6T and extensive 400G deployments spotlight network capacity as crucial for scaling AI.

The research highlights a transformation in industry thinking: it's no longer solely about infrastructure capacity but about optimising efficiency and reliability. As sophisticated AI models become mainstream, the importance of real-world AI workload emulation is underscored, offering a way to unlock infrastructure potential while managing costs.

"AI data centres are reaching a tipping point where performance and scale alone are not enough. Operators need deeper insight, tighter validation, and smarter infrastructure choices," explained Ram Periakaruppan, Vice President and General Manager, Network Applications & Security Group at Keysight, indicating the criticality of optimising networks in the AI era.