Energy demand for AI data centers in the U.S. is expected to grow about 50 gigawatt each year for the coming years, according to Aman Khan, CEO of International Business Consultants
In sum – what you need to know:
AI fuels infrastructure race – Hyperscalers like Microsoft, Google, Amazon and Meta have invested over $200 billion in AI infrastructure in 2024, with plans to exceed $220 billion in 2025, driving massive demand for power-hungry data centers.
Power becomes the bottleneck – AI workloads, especially training, require unprecedented electricity levels. Inference tasks like ChatGPT searches can use up to 30 times more energy than standard searches.
Energy innovation is critical – To meet growing demand, firms are optimizing power usage and exploring diverse sources including renewables, small nuclear reactors, and converted coal plants.
As AI adoption accelerates, the global tech ecosystem is scrambling to scale up its infrastructure. But powering this transformation comes with serious energy challenges, said Aman Khan, CEO of International Business Consultants, in a recent webinar organized by DatacenterDynamics.
Khan emphasized that AI models — particularly those used for training — require massive compute power, which translates into immense power demand. “As we have heard, the ChatGPT search may require 30 times more power to process a normal search in what power it needs in Google search engine,” he said. And that’s just inference workload, he noted, adding that training models demand even more power over extended periods.
He cited a projection for the U.S. market: “Power demand for AI data centers is expected to grow about 50 gigawatt each year for the coming years.” In Europe, he added, “every three to four years, this capacity will be doubling for the data center consumption.”
These demands are already reshaping investment decisions. “Investors are looking for AI training data center sites that are physically close to the power supply. If it is grid or renewable energy, power doesn’t matter,” Khan said. Other focus areas include high-performance computing, cooling systems and distributed AI models at edge data centers.
Khan also outlined major infrastructure investments globally: “All four hyperscalers —Microsoft, Google, Amazon and Meta — they have made Capex investments in AI infrastructure, more than $200 billion in 2024 alone.” He cited specific 2025 forecasts: “Meta has announced about $65 billion, and Alphabet or Google, $75 billion and Microsoft, about $80 billion.”
Adding to that, Khan mentioned the Stargate initiative in the U.S., “which is joined by SoftBank, Oracle and OpenAI, which is about a $500 billion project.”
In Europe, the European Union announced a 200 billion euro EU invest fund for AI-related projects. This fund will finance what they call four AI gigafactories across the EU.
Despite the enthusiasm, Khan flagged serious concerns about energy availability. “There is a major risk that some of the North American investors, they may focus on development of AI training data centers in North America or in Nordic rather than Europe, which will not be a good news for Europe.”
Asked whether new investments or a grid overhaul would be sufficient, Khan responded, “We need an overhaul. We need new approaches. We need both.”
He explained that companies are responding with two main strategies: improving energy efficiency and investing in alternative energy sources. “So if I’m sitting on 500 MW, my first focus is how can I use those 500 MW more effectively via monitoring, optimizing, initiating technologies, technology refresh projects, new solutions or even using AI?”
Alternative energy strategies include wind and solar parks, small modular nuclear reactors (SMRs), power purchase agreements (PPAs), and hydro/geothermal projects. But all come with hurdles. “The challenge we have when it comes to wind and solar parks is the energy storage, as wind and sun are not always there.”
Regarding SMRs, Khan noted, “There are about 20 projects in the U.S. and about the same in Europe,” but regulations are slowing progress. “We’re not talking about tomorrow. It’s going to take a while because of the permit process.”
Companies are also repurposing idle infrastructure. “Data center companies are sometimes evaluating the dependent coal plants or production facilities… trying to transform those abundant sites, abundant locations into data center facilities because the power is available.”
