The AI boom has transformed the tech landscape, but it's not just about the models themselves. It's about the physical infrastructure that powers them. Among the biggest winners in this boom is Caterpillar, a company known for its construction vehicles, but its success is now tied to its giant gas-powered engines that are crucial for the nation's data-center build-out. This shift highlights a fundamental change in Silicon Valley's advantage: it's no longer just about low-cost production, but about the heavy industry of data center construction and operation.
The AI industry's reliance on data centers is immense. These centers require tech companies to build power plants, set up electrical equipment, and manage complex cooling systems. This construction frenzy is approaching an inflection point, with major tech giants spending more on data centers than they earn from their operations. As a result, these companies are turning to debt to fund their AI ambitions, with capital expenditures exceeding half a trillion dollars in the past year alone. The trend is set to continue, with projections of similar spending in 2026 and over $1.1 trillion in AI investments next year.
The energy demands of data centers are staggering. A standard data center might require 10 or 50 megawatts of power, but now, data centers are demanding gigawatts of power. Meta's flagship AI data center, for instance, is being expanded to five gigawatts, and a proposed data center in Utah could demand nine gigawatts. This has led to a surge in demand for power equipment, with Caterpillar and other natural-gas-turbine manufacturers experiencing back orders. Elon Musk's recent purchase of an energy company with a fleet of combustion turbines is a testament to this growing need.
However, the challenges go beyond power. Data centers require intricate wiring and cooling systems to manage the heat generated by AI chips, which can reach up to 200 degrees Fahrenheit. This has led to a focus on HVAC efficiency, with executives like OpenAI's Sam Altman emphasizing the importance of 'electrons' (power) and skilled workers like 'plumbers and electricians'.
The AI industry's desperation to keep building is evident, but it's not without backlash. Anthropic, a safety-conscious firm, is spending $1 billion a month to rent a data center from Musk, who previously deemed them 'evil'. This highlights the growing opposition to data centers, with New York becoming the first state to institute a moratorium on new 'hyperscale' data-center construction. The AI backlash could escalate, as the industry grapples with the environmental and ethical implications of its rapid growth.
The AI supply chain is becoming a complex, global enterprise, with a focus on 'compute' and the minerals that feed it. The State Department's Pax Silica pact aims to bolster this supply chain, recognizing that the 21st century runs on compute and rare-earth minerals. Financial markets are treating 'tokens' (the basic unit of AI) as raw inputs to modern civilization, with firms launching markets for compute akin to those for oil, steel, and corn. This shift underscores the transformation of AI into a heavy industry, with Silicon Valley exerting indirect control over the physical world through its data-center build-out.
