Americans Finally Pulled Back at the Register

The artificial intelligence race is entering a new phase, and the biggest constraint may no longer be the quality of the models. It may be whether there is enough physical infrastructure to run them. Anthropic announced Monday that it is partnering with Macquarie Asset Management and Singapore’s sovereign wealth fund GIC to create a new infrastructure platform called Theseus Infrastructure. The venture will develop, operate, and lease large-scale data centers in the United States, with Anthropic serving as a long-term anchor customer. This partnership highlights how the AI industry is increasingly treating computing infrastructure as a strategic asset rather than something purchased from a cloud provider. The structure itself indicates that demand for AI computing has grown significantly, making financing, constructing, and securing data-center capacity a competitive factor. AI agents require enormous computing power because they perform complex tasks—reasoning through tasks, accessing company systems, executing multiple actions, reviewing work, and operating for extended periods. As businesses shift from asking AI questions to allowing AI systems to perform work, computing capacity consumption increases. Research on OpenAI’s Codex usage shows rapid growth in agentic AI adoption, with organizations increasingly using agents for complex workflows. This creates a fundamental problem: AI adoption can outpace the physical infrastructure required to support it. The infrastructure problem extends beyond servers and chips. Data centers consume vast amounts of electricity, and new facilities can strain regional power grids. PJM, the grid operator covering much of the eastern United States, warns that electricity costs could rise substantially over the next five years due to increased demand. Data-center operators also face risks of temporary disconnections to protect the grid. Anthropic will cover increased electricity costs for its new facilities, signaling a broader business issue: the AI economy is increasingly colliding with the physical economy. Companies need land, electricity, transmission infrastructure, cooling, advanced chips, construction capacity, and regulatory approvals. Even the best AI model can be constrained if it lacks sufficient computing capacity. For businesses, the immediate lesson is to recognize that AI infrastructure constraints will influence the cost, availability, and performance of AI services. Companies should avoid assuming that computing will always become cheaper and more abundant at the same pace. AI vendors may face increasing infrastructure costs, which could eventually appear in subscription prices, API pricing, usage limits, or enterprise contracts. Businesses should evaluate AI systems based on their ability to perform real work efficiently rather than just benchmark scores. The winning AI system may be the one that completes business processes reliably while consuming fewer resources. Anthropic’s infrastructure partnership underscores that AI is becoming an industrial business. The first phase of the AI boom focused on models, but the next phase is about deployment. Deployment requires physical infrastructure on a scale akin to telecommunications networks and cloud computing. This shift creates opportunities for data-center developers, utilities, power-generation companies, semiconductor manufacturers, networking firms, construction firms, cooling providers, and infrastructure investors. For ordinary businesses, the message is clear: AI is no longer just software. It is an increasingly massive physical system that businesses are becoming dependent upon. As AI moves from answering questions to performing real work, access to the infrastructure behind it may become one of the most important competitive advantages in the economy.