The adoption of artificial intelligence (AI) is accelerating faster than the pace of infrastructure being built, threatening to create bottlenecks in future workloads, according to a new report from Keysight Technologies Inc.
The report, “Beyond the Bottleneck: AI Cluster Networking Report 2025,” calls on telecoms and cloud vendors to shift from expansion to optimization to support next-generation AI workloads.
As AI models accelerate and become more sophisticated, tools like AI workload emulation are becoming critical for infrastructure while simultaneously controlling costs.
Of those surveyed for the report, 95% said real-world workload emulation is critical, but they lack the tools necessary to support production-scale AI environments effectively.
Lack of infrastructure
The lack in infrastructure is in the data center as AI accelerates across industries. But just building out more infrastructure is not enough according to most Keysight report respondents. 62% of those in the report said they aim to get more out of the infrastructure currently installed rather than invest in new infrastructure.
“AI data centers are reaching a tipping point where performance and scale alone are not enough. Operators need deeper insight, tighter validation, and smarter infrastructure choices,” said Ram Periakaruppan, VP and GM of the Network Applications & Security Group at Keysight. “This research confirms what we see in the field: Success in the AI era hinges on optimizing every layer of the network.”
The report said that operators are turning to performance optimization strategies such as:
- AI workload emulation
- Accelerate deployment of AI clusters
- Enhancing efficiency
Maintain or expand
Of those surveyed for Keysight’s report, 89% of respondents said they plan to expand or maintain AI infrastructure investments with the main areas of growth being:
- Cloud integration
- Faster GPUs
- High-speed network upgrades
However, about 62% of operators plan to extract more value from existing infrastructure without major new investments.
Adoption of high-speed networking is needed for AI adoption with many exploring 800G, 1.6T and Ultra Ethernet as options.
