Many businesses' infrastructures are not ready for AI.

Keysight Technologies, in collaboration with Heavy Reading, has released the Beyond the Bottleneck: AI Cluster Networking Report 2025. The report finds that artificial intelligence (AI) adoption is accelerating faster than infrastructure can support it. The global study highlights the urgent need for telecommunications and cloud service providers to shift from scaling to optimization to support next-generation AI applications.

62% of survey respondents said they wanted to make the most of existing infrastructure without new investments.

As AI deployments accelerate across industries, demand for data center infrastructure is also growing rapidly. Building new infrastructure is no longer enough, with 62% of survey respondents saying they want to get the most out of their existing infrastructure without making new investments. Operators are turning to performance optimization strategies, such as emulating real-world AI applications, not only to validate performance, but also to improve efficiency and accelerate the deployment of next-generation AI clusters.

Nearly 89% of survey respondents plan to increase or continue investing in AI infrastructure in the coming year. Reasons for this growth include cloud integration (51%), upgrading to faster GPUs (49%), and higher network speeds (45%).

Optimization-first approach: While continuing to invest, 62% of operators said they expect to extract more value from existing infrastructure without new investments.

Simulation becomes essential: 95% of survey respondents said simulation of real-world applications is a critical capability, but they still lack the tools to effectively simulate production-scale AI environments.

Growing pressure on infrastructure: More than 50% of operators consider budget constraints (59%), infrastructure constraints (55%), and talent shortages (51%) as the biggest obstacles to scaling AI.

The Rise of High-Speed Networking: Adoption of advanced networking technologies is accelerating, with 34% of respondents exploring 800G, 22% testing 1.6T, and 58% evaluating Ultra Ethernet as a high-performance networking option.

Addressing network bottlenecks is the next goal: With 55% of carriers deploying 400G links and exploring 1.6T technology, network capacity is seen as a determining factor in AI scalability.

Yellow River

Source: https://doanhnghiepvn.vn/chuyen-doi-so/co-so-ha-tang-cua-nhieu-doanh-nghiep-chua-san-sang-cho-ai/20250918065323163