A Decision Framework for Generative AI Deployments

Generative AI (GenAI) promises to deliver deep, actionable, and real-time insights in a conversational, human-friendly manner. By using proprietary data sets as inputs for GenAI algorithms, companies can transform every facet of their internal and external facets of their business, including productivity, competitiveness, and customer engagement. GenAI is not a fleeting trend; it is here to stay. Businesses must prepare for this disruption now.

An often misunderstood aspect of GenAI is that it is resource intensive. Many believe that GenAI initiatives require the development of massive "foundation models" (i.e., large language models [LLMs] with billions of parameters on accelerated computing instances in the public cloud). In fact, not all GenAI models are large. Similarly, not all organizations need to create models from scratch — for most it is overkill. These prevailing misconceptions lead to businesses making one or two assumptions, both of which can be expensive in the long run. First, that GenAI training or inferencing requires highly performant accelerated infrastructure, no matter how small or large the models. Second, public cloud is the only cost-effective way for accessing highly performant resources. Nothing could be further from the truth.