The artificial intelligence industry is approaching a critical inflection point in 2024, characterized by explosive growth in compute requirements and rapidly evolving market dynamics. Current analysis shows the industry significantly underestimating the infrastructure investments needed to support projected AI adoption, with 2025 requiring approximately $180B in chip-only capex – 251% above current consensus estimates. This gap between projected needs and current planning grows more dramatic through 2028, potentially reaching $1.7T in annual chip-only capex requirements.

Training compute demands are scaling at an unprecedented rate, particularly for frontier models. The next generation of leading AI models, such as GPT-5, will require approximately 2.1B exaFLOPs of capacity, necessitating nearly 3.7M specialized GPUs/TPUs for training in 2025 alone. The total infrastructure investment, including compute, interconnect, and supporting systems, could approach $200B annually. The progression of training compute requirements shows dramatic scaling: