As AI adoption increases, tech giants are investing hundreds of billions of dollars into large-scale data centers, i.e., vast warehouses filled with servers and specialized hardware such as GPUs. The podcast on Wired.com hosts Lauren Goode and Michael Calore speak with senior writer Molly Taft to unpack how these facilities actually operate, what they consume, and what risks the industry is facing.
They begin by breaking down the user journey: when you send a query to an AI service, it’s routed through authentication, moderation, and load-balancing before arriving at clusters of GPUs inside a data center. These GPUs process hundreds of millions of tokens in parallel to generate an answer in seconds.
The conversation then shifts to energy and environmental stakes. These data centers demand huge amounts of electricity, cooling, and infrastructure. Their carbon footprint depends heavily on whether the local grid is powered by fossil fuels or renewables. Some regions are already experiencing strain on utilities and grid stability.
Beyond energy, there are broader risks. The industry’s aggressive build-out assumes an ever-rising demand for compute. But if that demand does not materialize or efficiency improvements outpace hardware growth, we could see stranded assets or an “AI bubble.” There’s also a political dimension: local communities are pushing back on new data centers due to their power, water usage, and impact on utilities. Meanwhile, industry narratives and accounting practices around energy, cost, and demand remain opaque.
For engineers and tech-interested professionals, this episode offers a window into a critical infrastructure layer of our digital economy. It underscores that the AI surge isn’t just about algorithms and models; it’s grounded in kilowatts, water-cooling, regional grids, and industrial-scale build-out. If the foundation falters, the roof may too.
In short, the infrastructure behind AI is far from invisible, and the stakes of scaling it responsibly just got clearer.