Data Centres Are Now a Grid-Scale Energy Problem
The group representing electrical equipment manufacturers expects data centre energy consumption to grow 300% over the next 10 years. That forecast, published in the context of a 2026 IEEE conference focused on grid reliability, is not an aspirational projection by technology optimists — it is a supply planning number that electrical equipment manufacturers and grid operators are using to dimension the infrastructure investments required to serve it. The scale is difficult to contextualise against historical precedent. The electrification of the automobile fleet, the expansion of air conditioning in the American South, the post-war industrial build-out — each of these was a significant demand event for the power sector. The AI infrastructure build is, on current trajectory, larger than any of them, and it is happening faster.
The immediate commercial consequence is a fundamental restructuring of where and how power capacity is built. Some potential data centre customers are pausing decision-making in the PJM Interconnection to see how the grid operator's colocation and backstop auction rules shake out. The hesitation reflects a genuine uncertainty in the market: the regulatory frameworks governing how hyperscale data centre loads connect to and pay for the grid were designed for an era of slower, more predictable demand growth. Operators attempting to bring 500MW or 1GW campuses online in interconnection queues built for 50MW industrial customers are encountering process delays and cost uncertainty that are redirecting investment to more permissive jurisdictions — and in some cases, to off-grid or behind-the-meter power solutions that sidestep the interconnection bottleneck entirely.
The AI Stock Surge: Infrastructure Plays Outperforming the Broader Market
As of May 2026, AI infrastructure stocks like Bloom Energy, with a 130% revenue surge, and Sandisk, with 251% revenue growth, have more than doubled, outperforming the Nasdaq-100 and S&P 500, driven by data centre demand. The performance of Bloom Energy is particularly instructive for understanding the commercial dynamics of AI infrastructure investment. Bloom's fuel cell technology enables data centre operators to generate power on-site, independently of the electric grid — directly addressing the interconnection bottleneck that is constraining conventional grid-connected development. At a moment when grid connection timelines are measured in years rather than months, the ability to deploy behind-the-meter generation at data centre scale has moved from a niche offering to a mainstream infrastructure solution.
Sandisk's performance reveals a different but equally significant dimension of AI infrastructure demand. In its fiscal third quarter of 2026, ending April 3, Sandisk's revenue rose 251% year over year, fuelled by a 233% jump in data centre revenue, with operating income climbing 319% to exceed $4 billion. NAND flash memory — the core technology in Sandisk's data centre products — is being consumed at rates that reflect the fundamental architecture of large language model inference: massive, fast storage that can serve model weights and context windows to GPU clusters at the speeds required for real-time response. The companies supplying the unglamorous infrastructure components of AI — storage, power, cooling, networking — are in many cases outperforming the AI model developers themselves as commercial propositions.
The Grid Reliability Crisis: When AI Load Growth Meets Ageing Infrastructure
The North American Electric Reliability Corporation issued a Level 3 Essential Action Alert on May 4, urging grid operators to address risks from large data centre loads. A Level 3 Essential Action Alert is NERC's highest advisory category — the one reserved for risks that require immediate operational response rather than planning consideration. The fact that AI data centre load growth has risen to this level of grid reliability concern within a single year of peak hyperscale investment illustrates the pace at which commercial technology decisions are creating physical infrastructure consequences. The power grid was designed to handle the simultaneous failure of a single large facility. It was not designed to absorb 500MW of new load that was not in any forecast twelve months ago.
Rapid load growth from hyperscale data centres, the expansion of AI computing, and the rise of on-site generation needs are driving the largest shift in electricity demand in decades, with utilities, data centre operators, EPC firms, and technology providers all being pushed into new roles. The restructuring of roles is not merely organisational. It requires capital allocation decisions of a scale and speed that the utility sector has not managed since the original electrification of the American economy in the early twentieth century. Utilities that are moving fast — investing in new transmission, modernising interconnection processes, developing power purchase agreements designed for the load profile of AI data centres — are positioning themselves for a decade of revenue growth. Those that are not are watching investment redirect to competing jurisdictions.
The Investment Implications: Where the Money Is Moving
The commercial opportunity created by AI infrastructure energy demand is concentrated in four areas, each with distinct risk-return profiles. Grid infrastructure — transmission lines, substations, grid-scale battery storage — offers long-duration, regulated returns with low technology risk but high regulatory and permitting complexity. On-site generation — fuel cells, small modular reactors, gas turbines — offers higher margins and faster deployment but faces fuel cost exposure and evolving environmental regulation. Energy storage — both at grid scale and at the data centre campus level — is the fastest-growing segment, driven by the need to manage the intermittency of renewable generation and the peak demand profiles of GPU clusters. And the enabling technology layer — power electronics, cooling systems, monitoring software — is where the most disruptive margin expansion is occurring as incumbents compete with well-funded startups for data centre operator specification.