A new paradigm in artificial intelligence is emerging, shifting the focus from training advanced models to leveraging their capabilities through inference. In a significant trend noted for 2023, approximately two-thirds of AI computing power is projected to be allocated to operational use, marking a departure from earlier periods dominated by model training.
AI functions fundamentally through two processes: training and inference. Training involves extensive resources to develop models, while inference encompasses the utilitarian application of these models in decision-making tasks. For instance, when users interact with systems like ChatGPT, every query represents an inference request that demands computational resources. Deloitte’s analysis indicates that inference has rapidly evolved into the primary driver of AI’s computing power needs, growing from one-third last year to an anticipated two-thirds this year.
As companies increasingly implement AI agents, the demand for inference-driven computing is set to surge. The reliance on inference underscores the cost-effectiveness of decision-making processes, especially as these AI systems potentially make billions of decisions. Gartner’s estimates corroborate this trend, predicting global spending on AI infrastructure for inference will surpass that for training, reaching $23.3 billion compared to $19 billion.
A notable innovation in this arena is Jev, an AI designed specifically for decision-making without the iterative response generation typical of traditional models. This shift towards simpler, cost-effective tools reflects a broader industry transition focused on maximizing the efficiency of AI applications. The growing emphasis on inference signals a new chapter in AI development, highlighting the importance of operationalizing these technologies effectively.
- Why this story matters: It highlights a significant shift in AI processing, emphasizing the growing importance of operational applications over model training.
- Key takeaway: Inference is rapidly becoming the dominant use of AI computing power, influencing spending and operational strategies.
- Opposing viewpoint: Some experts may argue that the focus on inference could lead to neglecting advancements in training deeper, more complex AI models.
