AI data centers play a pivotal role in driving advancements in modern machine learning and computational technologies. However, a key challenge faced by these facilities lies in their substantial power consumption. Unlike traditional data centers that primarily handle storage and processing for standard business applications, AI data centers must support intensive workloads like deep learning, extensive data analytics, and real-time decision-making.
Furthermore, the report details a significant contrast between projected expectations and on-the-ground realities. While developers typically plan based on a 12 to 18-month timeline for grid power access, utility providers in major U.S. markets report potential extensions of up to two years, posing a substantial challenge in meeting the stringent deadlines essential for AI infrastructure deployments.
As AI-driven technologies continue to evolve, the rising energy intensity poses a pressing need for scalable power generation solutions. Bloom Energy’s report underscores that electricity access has now surpassed all other factors in the selection of data center sites.
According to Bloom Energy’s 2025 assessment, onsite power generation is poised to play a defining role in the forthcoming era of AI-powered infrastructure. The report also underscores the shift towards eco-friendly, rapid-deployment energy systems capable of managing the unpredictable energy demands inherent in large-scale AI operations.
Moreover, the report highlights that 95% of surveyed data center leaders have pledged to achieve carbon reduction goals. Yet, many acknowledge that the timeline for hitting these targets might need to be adjusted as a temporary focus shifts toward securing reliable energy sources.
Consequently, 84% of data center executives now prioritize power availability among their top three site selection criteria, overtaking considerations such as land costs or proximity to end-users, as revealed in the latest report.
The forecast indicates a dramatic expansion in the size of data centers, with median capacities projected to more than double from 175 MW to around 375 MW over the next decade. These facilities will necessitate more adaptable and reliable energy solutions, especially for AI-driven workloads that demand high-density computing.
Projections suggest that by 2030, 27% of data centers will be exclusively powered by onsite energy, showcasing a substantial increase from the mere 1% reported in 2024. This transition is primarily catalyzed by grid delays and the escalating energy requirements of AI technologies.
The lingering issue of grid delays has emerged as a critical aspect influencing decision-making processes. Utility companies are currently experiencing delays of up to two years in providing energy, which is significantly prolonging the selection of data center locations.
In a recent statement, Aman Joshi, the Chief Commercial Officer at Bloom Energy, emphasized the evolving landscape of data center site selection, pinpointing power accessibility as a pivotal factor. The escalating demands of AI applications have strained the grid’s capacity, prompting the industry to pivot towards self-generated power sources. The ability to exert control over power availability not only ensures timeliness but also differentiates viable projects from stagnant ones.
AI-driven workloads, particularly those reliant on deep learning and generative models such as GPT-4 and Google’s Gemini, necessitate immense computational power. The training of these models involves processing trillions of parameters, requiring thousands of high-performance GPUs or TPUs that consume significantly more power than traditional CPUs.
Recent findings indicate that nearly 27% of data centers are gearing up to rely solely on onsite generation by 2030, a substantial surge from the 1% recorded in 2024. An additional 11% are expected to heavily utilize onsite power. The upsurge in onsite power adoption is attributed to the escalating demands of AI workloads and ongoing delays in connecting to utility grids.
