Galaxy Digital (NASDAQ: GLXY) has pointed out in a blog post that in an era where artificial intelligence is reshaping every aspect of human life—from work and entertainment to commerce and creativity—the fundamental question now arises: Do we have sufficient power to fuel this digital transformation?
A recent perspective from Galaxy Interactive, a venture arm focused on interactive technologies, delves into this critical issue, highlighting energy as the ultimate limiter for AI’s expansion.
Galaxy Digital’s insights underscore how compute, once seen as boundless in the cloud, is inherently tied to physical infrastructure, with electricity emerging as the primary bottleneck.
The core argument revolves around the merging of digital and physical worlds.
As AI-driven systems like spatial computing and persistent virtual environments become ubiquitous, the demand for computational power skyrockets.
However, this isn’t just about algorithms or chips; it’s about the energy required to run massive data centers.
Globally, electricity supply seems adequate on paper, but in high-demand hubs such as Northern Virginia, Dallas, and Phoenix, grids are strained.
Substations are maxed out, transformer shortages cause years-long delays, and transmission lines can’t keep pace.
This mismatch means that even cutting-edge AI models could stall not due to technological limits, but because of insufficient power delivery.Supporting this view are stark projections.
Data centers worldwide currently consume around 47 gigawatts (GW), about 1.5% of global electricity, but this could surge to 111 GW by 2030, equating to nearly 1,000 terawatt-hours annually.
In the U.S. alone, usage might jump from 20 GW (4% of national power) to 100-130 GW (over 15%) in the same timeframe—a growth rate five to ten times faster than the grid’s historical expansion.
For context, a single AI campus might require 300-800 megawatts (MW), enough to power hundreds of thousands of homes, while mega-projects like a proposed 5 GW facility in the UAE rival the output of several nuclear plants.
These figures don’t even account for overheads like cooling and efficiency losses, which inflate actual needs by 30-50%.
To illustrate the scale, consider everyday AI interactions: A simple web search uses about 0.3 watt-hours, equivalent to pedaling a bike for a few seconds, but a multi-step AI task could demand 1 kilowatt-hour—hours of intense cycling.
As AI training mimics steady loads like cryptocurrency mining and inference varies like web traffic, the energy profile shifts, concentrating stress on already congested areas.
Galaxy Interactive proposes solutions centered on nuclear energy for its reliability, high density, and scalability.
They’ve invested in fusion technology through Commonwealth Fusion Systems, eyeing long-term abundant clean power, and in modular fission via Last Energy for quicker deployment.
Nuclear offers advantages like a 90% capacity factor—far above renewables—and compact designs suitable for urban edges.
Industry momentum is building: Tech giants like Google and Meta are securing nuclear deals for gigawatts of capacity, while policies provide tax incentives and regulatory streamlining.
Yet challenges persist. Building the required 50-80 GW of firm U.S. power by 2030 demands aggressive action, potentially blending nuclear (40%), solar (40%), gas, and storage.
Efficiency innovations, such as edge computing or in-memory processing, can help mitigate waste but won’t eliminate the need for infrastructure growth.
Underestimations in models—ignoring real-world overheads—could exacerbate shortages, with experts warning that power constraints might cap AI progress within years.
Ultimately, this energy thesis signals a paradigm shift: Scaling intelligence means scaling power first. As Galaxy expands operations, like their 1.6 GW-approved mining campus, they position themselves at the intersection of tech and energy. The message is seemingly clear—without substantial investments in reliable sources like nuclear, the AI future risks running out of juice before it fully ignites.