The numbers coming out of Goldman Sachs research are stark. U.S. data center power demand is set to more than double in two years, jumping from 31 gigawatts in 2025 to 66 gigawatts by 2027. The primary driver is the buildout of AI infrastructure. This is not a distant problem. It is happening now.
The math is simple. AI workloads currently account for 14% of global data center demand. That figure is expected to hit 27%. As that share rises, so does the energy appetite. The environmental impact is already significant. It will get worse before it gets better.
Many organizations have signed sustainability commitments. Those pledges often sit in direct conflict with their actual AI usage. A company cannot promise net-zero emissions while doubling down on power-hungry large language models. The contradiction is plain. The report makes clear that these two goals do not align.
This is where the conversation about edge computing enters the picture. The technology processes data closer to where it is generated, rather than shipping everything to a central data center. That reduces transmission losses. It also allows for smaller, more targeted computing footprints. For certain AI tasks, edge computing can deliver results without the massive energy overhead of a centralized facility.
But edge computing is not a silver bullet. It works best for specific, low-latency applications. Training large AI models still requires the raw computational muscle of a data center. The real question is how much of the AI workload can shift to the edge. If the answer is a significant portion, the pressure on data center power demand could ease. If not, the projected 66-gigawatt figure may prove conservative.
The forces behind this are structural. AI is not a fad. It is being embedded into everything from customer service to medical diagnostics. The demand for more powerful models is relentless. Each new generation of AI requires more compute. That means more data centers. That means more power.
Goldman Sachs projects the power demand climb from 31 to 66 gigawatts in just two years. That is a 113% increase. For context, 66 gigawatts is roughly the output of 60 large nuclear reactors. Building that much new generation capacity in two years is a monumental challenge. It is not clear where the power will come from.
The logical path forward is a mix of approaches. Efficiency improvements in hardware. Better cooling technologies. Greater use of renewable energy. And yes, edge computing for the workloads that can handle it. But none of these are quick fixes. The infrastructure decisions made today will lock in energy consumption patterns for decades.
What happens if the power is not there? Data center buildout slows. AI deployment stalls. The economic projections that depend on AI growth start to look shaky. That is the risk. The report does not say this explicitly, but the implication is clear. The sustainability problem is not just an environmental issue. It is a bottleneck for the entire AI industry.
Organizations face a hard choice. They can honor their sustainability commitments and constrain AI growth. Or they can prioritize AI and watch their carbon footprint balloon. The report notes the misalignment. It does not offer an easy way out.
The next two years will test how serious these sustainability pledges really are. If data center power demand hits 66 gigawatts, the gap between rhetoric and reality will be measured in megawatts. Edge computing might help bridge that gap. But it is not a panacea. It is one tool among many. The industry needs to use all of them. Fast.
