Building the Next Generation of AI Infrastructure

Chapter 1. Long Live the AI Infrastructure

The notion of AI infrastructure is not new, but it is becoming increasingly relevant for organizations around the world as they scale their level of AI adoption. Even in the era of pretrained GenAI, organizations still need solid platforms and specialized hardware. The lack of a strong architectural foundation is a major bottleneck for companies that don’t own their own bare metal environments. Additionally, the nature of the infrastructure changes over time, as it is not the same to train new machine learning (ML) or deep learning (DL) models, fine-tune existing large language models (LLMs), or move to production and inference your final AI deployments.

And the role of AI infrastructure is relevant even in times of a clear switch from only science activities to a mix of science and engineering to bring AI-enabled applications at scale. That mix keeps the best of both worlds, but it also includes considerations like experimentation being the norm in terms of new product creation, and scalability (whatever you want that to mean, depending on your context) to connect existing resources in a pragmatic, efficient, and powerful way. As an organic consequence of that, the nature of day-to-day activities has now changed. What before was about preparing and mining data sources, today is—for most organizations—an exercise in technical administration and API-enabled AI capability deployments that switch the attention to the classic development ...