Idaho Is a Demand Story, Not an Energy Story
The announcement that next-generation nuclear reactors are moving toward construction at Idaho National Laboratory is being read as an energy infrastructure story. It is not. It is a demand story. Artificial intelligence inference workloads have created an electricity appetite that existing grid capacity struggles to meet at baseload reliability. The nuclear investment is downstream of a single variable: the continuous compute requirements of large language models running at commercial scale. When the International Energy Agency modeled the trajectory in its January 2025 electricity outlook, the numbers made the implied infrastructure build visible.
Hyperscalers Are Signing Generation Contracts, Not Renting Compute
Microsoft's decision to restart Three Mile Island Unit 1 through a long-term power purchase agreement with Constellation Energy made the economics visible at a single transaction. The deal committed 835 megawatts of baseload nuclear power exclusively to Microsoft's data center operations — power sold before the reactor was even restarted. Google followed with a series of small modular reactor agreements with Kairos Power. Amazon Web Services structured nuclear capacity acquisition through Talen Energy. These are not renewable energy credits or carbon offset line items. These are generation contracts — the same category of infrastructure commitment that utilities make when breaking ground on new plants. The capital signal is clear: AI infrastructure has become an energy procurement discipline, one that now sits on the same balance sheet as cloud spend and data center lease obligations.
The Application Layer Is Inside the Energy Budget
The connection to web framework architecture is not abstract. Every server-rendered page request consumes compute. At the scale of tens of millions of requests per day — standard for an enterprise web property — aggregate compute load is measurable in infrastructure cost terms. A Next.js application using Incremental Static Regeneration serves the majority of requests from CDN edge cache nodes, with server-side execution occurring only at cache invalidation intervals. The origin compute requirement for cached traffic approaches zero between build cycles. A WordPress installation executes a PHP process and at least one MySQL query for every uncached request, with active plugins compounding the query load at each layer. Across WebPulse's 466,000-site scan, WordPress accounts for 74% of detected CMS deployments — a compute model that was designed for a traffic paradigm that predates both AI inference workloads and the energy pricing pressures now reshaping hyperscaler procurement.
What This Means for Technology Budget Holders
The nuclear revival signals that AI energy costs are being priced at the infrastructure layer, not absorbed as a variable overhead externality. Organizations that have deferred framework modernization on the grounds that legacy CMS deployments are operationally stable are carrying a compute model whose energy signature is now quantifiable against a rising cost curve. AI agents — which constitute a growing and measurable share of inbound web traffic across WebPulse's detected sites — do not benefit from dynamic server-side rendering the way human browser sessions do. They fetch structured content; session-based personalization and dynamic assembly deliver no incremental value to machine consumers. The compute overhead of legacy CMS architecture absorbs energy cost while producing no additional output for the traffic category that is growing fastest. Framework architecture decisions that optimize for edge delivery, static generation, and reduced origin compute are, in the current energy pricing environment, infrastructure cost decisions as much as they are engineering choices. The framing that places those decisions exclusively in an engineering backlog is becoming harder to justify at the capital planning level.
