The Energy Limits of Artificial Intelligence, Without Fossil Fuels, By Brian Simpson and James Reed
The Zero Hedge article from March 4, 2025,
highlights a looming energy challenge in Texas, where the Electric Reliability Council of Texas (ERCOT) projects that the state's power grid will require the equivalent of 30 nuclear reactors' worth of electricity by 2030 to meet the surging demand from data centres powering artificial intelligence (AI) and other technologies. While the piece focuses primarily on infrastructure and energy capacity, it implicitly raises critical human resources (HR) issues tied to this rapid expansion. One key HR concern is the workforce required to build, operate, and maintain this massive energy infrastructure. Agee Springer, ERCOT's senior manager of grid interconnections, notes the unprecedented scale of industrial loads threatening grid reliability, a shift that demands a skilled labour pool far beyond current capabilities. The article does not directly address HR specifics, but the scale of the challenge—adding capacity equivalent to 30 nuclear plants in five years—suggests a need for thousands of engineers, technicians, and construction workers, many of whom must be trained in specialised fields like nuclear technology or grid management. Texas, already a hub for energy production, faces a labour market stretched thin by existing oil, gas, and renewable projects, making recruitment and retention a pressing issue.
Another HR-related implication is the cost and allocation of resources to train this workforce. The article references supply chain bottlenecks for equipment like turbines and transformers, but a parallel bottleneck exists in human capital. Nuclear projects, in particular, require rigorous safety training and certification, processes that can take years and strain educational institutions and corporate HR departments. Additionally, the piece hints at a cost-shifting debate via Texas Senator Charles Schwertner's critique of the "four coincident peaks" (4CP) program, which allows large industrial users like data centres to avoid grid upgrade costs by reducing peak usage. This flexibility shifts financial burdens onto households and smaller businesses, potentially sparking public backlash that HR teams must manage—think employee morale in energy firms facing public scrutiny or union pressure for better wages to offset rising living costs. Finally, the article's mention of data centres' profitability until power prices hit $2,000 per megawatt-hour (versus ERCOT's $5,000 cap) underscores a tension between corporate profits and workforce compensation, a classic HR challenge in balancing stakeholder interests.
Turning to the broader case for the energy limits of the information society, the Texas scenario is a microcosm of a global predicament. The information society—driven by AI, cloud computing, and data centres—relies on an insatiable appetite for electricity that is pushing physical and economic boundaries. A Bloomberg report from February 28, 2025, corroborates ERCOT's forecast, noting that Texas's peak power demand could jump 75 percent by 2030 from its current 85.5 gigawatts, driven largely by AI data centers. This aligns with a December 6, 2024, Zero Hedge piece estimating that data centres will consume 8 percent of U.S. power by 2030, up from 3 percent in 2022, outpacing renewable energy growth. The International Energy Agency (IEA) warned in October 2024 that this surge threatens climate goals unless tech firms pivot to clean energy, a transition lagging behind demand. For context, a single gigawatt—enough to power 250,000 Texas homes—is now routinely requested by individual data centre projects, per ERCOT's Springer. Globally, the IEA projects electricity demand doubling by 2040, with tech driving much of it, yet renewable deployment remains bottlenecked by land, materials, and grid upgrades.
The energy limits become stark when considering physics and economics. Nuclear power, touted as a solution, faces long lead times—small modular reactors (SMRs) won't be widely commercial until the 2030s, per a September 12, 2024, Zero Hedge report on Oracle's plans. Renewables like solar and wind, while growing (12 gigawatts added in the U.S. in 2024's first half, per an August 23, 2024, Zero Hedge article), are intermittent and land-intensive, incapable of matching AI's 24/7 hunger. Fossil fuels, particularly natural gas, fill the gap—Dominion Energy in Virginia expects an 85 percent demand rise over 15 years, per an October 16, 2024, ZeroHedge piece on Amazon's SMR investments—but this undermines decarbonisation pledges. Economically, the cost of scaling energy infrastructure is astronomical. Beth Garza of the R Street Institute warns, "There can't be more demand than supply," yet supply chains for transformers and skilled workers are already strained. A February 28, 2025, POWER Magazine article on Last Energy's 30-microreactor plan in Texas notes regulatory and financing hurdles, suggesting even innovative solutions lag behind need.
The information society's growth is thus self-limiting. AI's exponential computational demands—doubling every 100 days, per NVIDIA's Jensen Huang—outstrip Moore's Law and energy availability. If Texas, the U.S.'s energy titan, needs 30 reactors' worth of power in five years, imagine the global scale: hundreds of gigawatts, trillions in investment, and a workforce that doesn't yet exist. The HR issues—training, recruiting, and managing labour for this buildout—amplify the constraint. Without a radical energy breakthrough or a slowdown in tech expansion, the information society risks hitting a wall where data's promise exceeds the planet's power to sustain it. Texas's plight is a warning: the digital future is tethered to an analog reality of watts and workers, and both are in short supply.
The only solution here is to abandon Green/environmentalist mythologies, and to continue to use fossil fuels, especially coal. Technological innovations will work to diminish the pollution produced.
Comments