The rise of artificial intelligence isn’t just changing how we build software or run businesses, it’s reshaping the physical infrastructure that supports it. According to a recent analysis, data centers will consume one-fifth of U.S. electricity by 2035, up from current levels. That’s a staggering leap, and it’s not just about servers or cooling systems. It’s about the sheer volume of compute power AI is demanding, nearly 200 gigawatts by 2035, according to forecasts from BloombergNEF, EPRI, and S&P.
This surge isn’t evenly distributed. The U.S. will host 64% of global AI chip power demand by 2033, meaning the burden falls disproportionately on American grids. Regional systems like PJM and ERCOT are already feeling the pressure, with 34% and 22% of their capacity, respectively, expected to be allocated to data centers by 2035. That’s not a minor inconvenience. It’s a systemic risk.
The implications for enterprises are immediate. If your AI operations are scaling, you’re not just buying more GPUs or cloud instances, you’re committing to a physical infrastructure that must be resilient, scalable, and strategically located. Grid constraints could force delays, increase costs, or even cause outages if not addressed proactively.
One of the most interesting angles here is how this affects the very nature of AI deployment. If AI compute is becoming a grid-level resource, then the way we think about infrastructure, from power sourcing to geographic placement, must evolve. Companies that ignore this will find themselves stuck with legacy systems that can’t keep up with demand. Those that plan ahead will be able to scale without interruption.
This isn’t just a technical challenge, it’s a strategic one. Enterprises need to start thinking about energy sourcing now. Are you relying on grid power? Are you considering renewable energy partnerships? Are you designing your AI infrastructure to be modular and relocatable? The answers to these questions will determine whether your AI initiatives succeed or stall.
The good news is that this trend is also driving innovation. As grids strain, new solutions are emerging, from AI-driven energy management systems to decentralized compute architectures that reduce reliance on centralized power. Some companies are even exploring on-site renewable generation to power their AI workloads. This is not just about cost, it’s about control and resilience.
It’s also worth noting that this forecast has been revised upward by major consultancies. That means the problem is not theoretical, it’s already being felt in the market. The fact that BloombergNEF, EPRI, and S&P are all adjusting their projections upward suggests that the industry is moving faster than previously anticipated. This is not a distant concern, it’s a near-term reality.
For enterprises, the takeaway is clear: AI infrastructure planning must include energy strategy. That means not just buying more compute, but understanding how that compute will be powered. It means working with grid operators, exploring renewable energy partnerships, and designing systems that can adapt to changing power availability.
This is not just about avoiding outages, it’s about ensuring that your AI initiatives remain competitive. If your infrastructure can’t keep up with your AI ambitions, you’ll fall behind. The companies that succeed will be those that treat energy as a core component of their AI strategy, not an afterthought.
As AI adoption accelerates, the physical infrastructure supporting it will become more critical than ever. The data centers of the future won’t just be places where code runs, they’ll be energy-intensive ecosystems that demand careful planning, strategic partnerships, and grid-level resilience. The companies that get this right will be the ones that thrive in the AI era.
This forecast was first reported by TechCrunch. For more on how AI hardware is being financed, check out our post on GPU Financiers Bet Big on Inference Chips for AI Deployment.