Big quote: Energy supply has become the defining challenge in AI infrastructure, shaping innovation faster than hardware availability. Microsoft CEO Satya Nadella highlighted this shift during a joint interview with OpenAI CEO Sam Altman, explaining that the main obstacle to deploying large-scale AI today is the lack of power needed to run advanced GPUs. "The biggest issue we are now having is not a compute glut, but it's power – it's sort of the ability to get the builds done fast enough close to power," he said.
Nadella, who shared these insights on the Bg2 Pod YouTube channel, revealed that Microsoft has encountered situations in which hardware inventory exceeds what its data centers can actively support, naming electricity as the critical missing link.
The ripple effects of AI's energy demand are now visible at multiple levels. Modern data centers supporting large AI models can require as much electricity as a small city. Some hyperscalers under construction will use 20 times more power than existing sites, with individual campuses projected to need up to 2 gigawatts – a figure that rivals the total power demand of some US states.
According to recent estimates, US data centers consumed 183 terawatt-hours of electricity in 2024, equal to over 4% of total national power use, and this is expected to more than double by 2030. By 2028, AI-specific tasks alone could use as much energy as 22% of US households.
Data center operators now face increasing pressure to secure so-called warm shells – facilities equipped with the necessary utilities for immediate hardware installation. Because these sites must have adequate electricity and cooling capacity in place before new compute resources can go online, cloud providers and AI firms are often left with servers sitting idle for months as they wait for regional power constraints to be resolved.
Nadella described the issue bluntly: "[Y]ou may actually have a bunch of chips sitting in inventory that I can't plug in. In fact, that is my problem today. It's not a supply issue of chips; it's actually the fact that I don't have warm shells to plug into."
Since the resolution of the global GPU shortage, commercial operations have contributed to upticks in residential power bills, with some US states reporting increases of up to 36%. OpenAI has publicly called for federal investments in new power generation, arguing that energy has become a strategic asset in the global race for artificial intelligence. Altman and others regard China's investments in hydropower and nuclear energy as a significant advantage in scaling AI infrastructure, warning that current US capacity lags behind future requirements.
The intense power draw of AI also drives surging water demand, as many facilities rely on water-intensive or advanced liquid-cooling systems to keep servers and GPUs operating efficiently. In regions already struggling with water shortages, this has forced some data center operators to relocate projects to areas with naturally cooler climates to mitigate both environmental and cost risks.
Additionally, as data centers chase faster, more reliable energy, tech companies are moving aggressively to secure deals with utilities for priority access, resulting in overlapping projects and further strains on local grids.
Now, speculation is mounting that consumer hardware could soon run advanced AI models like GPT-5 or GPT-6 locally at very low power. Nadella and Altman suggested that as semiconductor technology improves, these low-power devices could eventually reduce the demand for ever-larger centralized data centers. That prospect has prompted questions among investors weighing the risks of multi-billion-dollar infrastructure projects against the pace of chip and device advancements.