
The next phase of artificial intelligence hinges on the physical constraints of power and silicon, specifically the availability of watts and wafers. Gavin Baker, CIO of Atreides Management, argues that while capitalism is effectively addressing energy shortages, Taiwan Semiconductor’s capacity decisions remain the primary variable for avoiding an AI market bubble. The industry is shifting from flat-rate pricing to usage-based models, which accelerates revenue growth for frontier labs. Meanwhile, the disaggregation of AI workloads into pre-fill and decode phases allows for more specialized chip architectures, extending the useful life of existing compute infrastructure. Despite the rapid pace of innovation, economic value remains concentrated at the frontier model layer, creating a high-stakes environment where companies must navigate geopolitical risks, cybersecurity threats, and the necessity of developing specialized, hard-to-replicate hardware solutions to maintain a competitive edge.
Sign in to continue reading, translating and more.
Continue