AI Agents Are Thirsty for Power
Silicon Valley is shifting away from chatbot queries toward a future filled with resource-intensive agentic AI—and it's driving the data center buildout.
By Molly Taft • Science • September 13, 2026
For years, the public conversation around artificial intelligence revolved around the chatbot: a single prompt, a single response, a relatively modest slice of computing power. That era is rapidly giving way to something far more demanding. AI agents—systems capable of planning, reasoning, and executing multi-step tasks autonomously—are becoming the central focus of the industry's biggest players. And they are hungry.
Unlike a chatbot query, which typically resolves in a single inference pass, an agentic workflow may involve dozens or even hundreds of sequential model calls, tool invocations, and self-corrections before completing a task. Each of those steps consumes energy. Multiply that across millions of users and thousands of enterprise deployments, and the result is a staggering new baseline for global electricity demand.
From Queries to Workflows
The distinction matters. A traditional chatbot interaction is bounded: the model receives an input and produces an output. An AI agent, by contrast, operates more like a digital worker. Given a goal—book a trip, debug a codebase, negotiate a contract—it breaks the objective into subtasks, calls external tools, evaluates its own progress, and iterates until the job is done. That autonomy is precisely what makes agents useful. It is also what makes them power-intensive.
Industry analysts note that a single agentic task can consume anywhere from ten to a hundred times the energy of a comparable chatbot exchange, depending on complexity and the number of reasoning steps involved. As enterprises rush to deploy agents across customer service, software development, logistics, and finance, the aggregate load on data centers is climbing sharply.
The Data Center Buildout
The infrastructure response has been swift and enormous. Hyperscalers and AI-focused cloud providers are racing to construct new data centers at a pace not seen since the earliest days of the commercial internet. These facilities are no longer concentrated in a handful of traditional hubs; they are spreading into new regions chosen for cheap land, favorable tax treatment, and—critically—access to abundant power.
Yet power is increasingly the bottleneck. Grid operators in several regions have warned that demand from new data centers is outpacing their ability to bring generation and transmission online. In some cases, utilities have had to delay retirement of fossil fuel plants or fast-track natural gas projects to meet the load. Nuclear energy, long out of favor, has re-entered the conversation as a potential long-term solution, with several major technology companies signing power purchase agreements and investing directly in advanced reactor development.
A Leadership Obsessed with Scale
The urgency is visible in the rhetoric of Silicon Valley's most prominent figures. Sam Altman and other leaders have repeatedly framed energy as the defining constraint on AI progress, arguing that the path to more capable systems runs directly through the power grid. Their message is unambiguous: without a massive expansion of energy infrastructure, the agentic future simply cannot be built.
That framing has consequences beyond the technology sector. It reshapes investment priorities, influences energy policy debates, and raises hard questions about sustainability. Data centers already account for a growing share of global electricity consumption, and the agentic shift threatens to accelerate that trend well beyond earlier forecasts.
What Comes Next
The tension between AI ambition and energy reality is unlikely to resolve quickly. On one side, the economic incentives to deploy agents are powerful, and the capabilities they unlock are genuinely transformative. On the other, the physical limits of power generation, transmission, and cooling are not easily overcome.
What is clear is that the chatbot era's relatively modest footprint was something of a prelude. The agentic era is arriving with a far larger appetite—and the world is only beginning to reckon with what it will take to feed it.
Image: Sam Altman. Credit: Wired
via Wired AI