Building data centre networks for AI workloads: key requirements and considerations for CSPs
11 October 2023 | Research
Strategy report | PPTX and PDF (6 slides) | Cloud and AI Infrastructure
As communications service providers (CSPs) train continuously larger AI models (especially generative AI models), models must be trained on an increasingly large number of graphics processing units (GPUs). CSPs that want to train these models in their own data centres will need to upgrade their back-end networks, that is, the networks which connect different GPUs within a data centre.
Information included in this report
- Insights into the back-end data centre networking requirements of AI workloads
- Analysis of the approaches to AI networking that CSPs can adopt in their data centres
Author

Joseph Attwood
AnalystRelated items
Case studies report
GPU-as-a-service (GPUaaS): telecoms operator case studies and analysis
Article
Should operators embrace cloud-native automation or evolve NFV MANO for autonomous networks?
Strategy report
Cloud-native automation and ETSI NFV MANO: evolution, migration and co-existence strategies