If Wall Street is the thermometer of what really matters, the fever for energy IPOs leaves no doubt: artificial intelligence has a new master, and it’s electricity. In the first half of the year, energy companies raised $12.6 billion from stock market listings, the highest half-year tally since the dot-com bubble peak in 1999 and well above the $4.3 billion collected in all of the previous year. Driving the stampede is not oil prices or renewables alone, but the insatiable appetite of data centers that train and run inference on Large Language Models.

The Dealogic figures signal a radical shift in perception. For years, energy was a taken-for-granted commodity, a variable cost to optimize with forward contracts. Today it becomes the critical raw material of a multi-trillion-dollar supply chain. Large funds aren’t seeking exposure to copper or wind farms out of environmental virtue: they want a ticket to the AI boom, an industry that could absorb between 10% and 20% of global electricity by 2030. And whoever controls generation and transmission suddenly finds themselves with unprecedented bargaining power.

The on-premise ripple effect

For the AI-RADAR crowd — those who design on-premise deployments, assess Total Cost of Ownership (TCO) of GPU clusters, or plan air-gapped installations — the news isn’t just financial. Energy already accounts for up to 40–50% of a modern data center’s operating cost. In a world where each NVIDIA H100 rack can draw more than 10 kW, choosing where to place servers becomes a matter of competitive survival. Geographic areas with abundant electricity and long-term contracts — often tied to renewables or nuclear — are emerging as de facto hubs for on-premise AI, while regions with aging grids or volatile prices risk being left behind.

The IPO wave isn’t limited to traditional utilities. Many of the newly listed companies develop microgrids, battery storage, and plants dedicated to tech campuses. This means that an operator who stays pure cloud today pays not just a compute tariff, but an implicit premium for the energy security that hyperscalers are locking in through producer contracts. Over the long term, those who own or control their supply — for instance via a small on-site gas plant or a dedicated solar farm — could gain such a cost advantage that self-hosting becomes the rational choice, even for midsize enterprises.

The end of plug neutrality

There’s a second-order effect few are focusing on. So far, the on-premise debate has centered on privacy, latency, and data sovereignty. The energy IPO frenzy introduces a new parameter: resilience of the electrical supply chain. A cyberattack on a power provider or a drought reducing hydro output can halt a fine-tuning operation as effectively as a software bug. In this scenario, distributed architectures spanning multiple sites with independent sources (edge computing with dedicated panels?) might stop being science fiction.

Markets are already pricing in this complexity. It’s no surprise that large private equity funds are buying land adjacent to high-capacity grid nodes and signing decade-long power purchase agreements even before building the sheds. Latecomers will have to settle for crumbs or pay extortionate prices.

AI ate software, as the saying went a few years ago. Now it’s devouring cables, turbines, and electrical substations. And the IPOs of these months are the birth certificate of a new order: the real bottleneck isn’t in GPU cores, but in megawatts.