The Digital World’s Hidden Cost: Rethinking Energy in the Age of AI

The Energy Behind the Internet 

The internet is physical. It relies on vast data centers, sprawling server farms, global networking infrastructure, and billions of connected devices. All of it runs on electricity. As digital tools have become more integral to everyday life, electricity usage tied to internet activity has skyrocketed. The generation of this electricity, primarily via coal and natural gas power plants, accounts for over a third of all energy consumption worldwide (the largest single source) as well as the CO₂ emissions that result. While renewable energy sources continue to gain a larger place in the global energy production mix, polluting fossil fuels still account for nearly 60% of the total, with more and more demand as power needs grow. 

Global trends: Over the last twenty years, electricity consumption from internet infrastructure has ballooned. Today, estimates suggest the Information and Communication Technology sector accounts for 6-12% of global energy use and 2–4% of global carbon emissions—comparable to the aviation industry. Data centers’ worldwide electricity consumption in 2026 is projected to reach 1,000 terawatts, roughly Japan’s total consumption.  

According to a report by the Berkeley Lab, electricity consumption by U.S. data centers rose from 58 terawatt-hours (TWh) in 2014 to 176 TWh in 2023, with projections of between 325 and 580 TWh by 2028 or nearly 12% of the country's total electricity. To put this into perspective, a typical large-scale power plant (coal, natural gas, nuclear) produces less than 10 TWh per year. Wind and solar power plants typically produce less than 2 TWh.   

It’s growing: With increased cloud adoption, streaming, and real-time AI tools, projections show that internet electricity use could double by 2030.  

AI: The New Energy Giant 

AI is revolutionizing industries—but it’s also energy-intensive. Training large models like Open AI's ChatGPT, Google's Gemini, or Microsoft's Copilot requires vast computational resources, often running for weeks or months in high-performance data centers.   

Training a single large language model can emit as much CO₂ as five cars over their entire lifetimes. And that’s just training—ongoing usage (aka inference) also draws power at scale. ChatGPT consumes over 500,000 kilowatt-hours (kWh) daily to process approximately 200 million requests. This daily consumption is about 17,000 times that of an average U.S. household, which uses around 29 kWh per day.  

Competing models (e.g., DeepSeek, Claude, Gemini) mean more computing demands and more data centers. DeepSeek claims to provide a much more energy efficient model (50-75% less intensive than ChatGPT). However, analysts indicate this efficiency is found in the model's training phase while inference (i.e., daily usage) may carry considerably higher energy usage due to more complex computing techniques and longer responses, largely negating the initial savings. Regardless, as AI adoption becomes widespread across sectors, electricity demand from AI alone could rival that of entire countries within a decade. 

Project Stargate 

OpenAI and Microsoft’s reported "Stargate" initiative seeks to invest $500 billion over the next several years to build the next generation of hyperscale data centers needed to run the growing AI infrastructure. However, estimates show that the current US electricity grid may be insufficient to power this additional demand, causing the need for large scale power plant construction. While Stargate has indicated this will include sustainable and environmentally friendly energy solutions (wind, solar, nuclear, carbon capture technologies, etc.) it remains unclear what the total impact will be on CO₂ emissions. But it’s hard to imagine a situation where increased fossil fuel burning is not part of the equation. 

Websites: Small Pages, Big Impact 

It’s not just AI. Even individual websites carry an energy cost, and when scaled across billions of users, those small bits of consumption matter. File size, media-heavy content, inefficient code, and bloated frameworks are major contributors. Every unnecessary JavaScript file or oversized image adds to power demand. 

A single person’s daily digital activity—browsing, streaming, emailing—can add up to several kilograms of CO₂ emissions per year. While this is difficult to quantify with every person's digital activity across the billions of interactions with the internet, the Sustainable Web Design Model provides the most in-depth and comprehensive formula used to determine energy consumption and corresponding emissions averages. The result? The average web page emits around 1.76g of CO₂ per view. Multiply that by thousands, millions, or billions of pageviews, and you start to see the impact of web performance on the environment.  

Want to check how your website rates? Try the Website Carbon Calculator to see your grade.

Building a Greener Web 

The good news? We can build better. Sustainable web design is a concept that puts people and the planet ahead of excessive media, graphically intensive experiences with little to no value for the audience, and unnecessary page bloat. The goal is to design web pages that drastically lower the average CO₂ emissions per page view by emphasizing efficiency, renewability, and a deeper understanding of the true impact digital products have on the world. Designing and developing with sustainability in mind not only reduces emissions—it often improves speed, UX, and long-term maintainability as well. We can also make smart choices about which products we use, emphasizing the environmentally responsible organizations that are striving to be part of the solution, not the problem.  

• Code smarter: Prioritize lightweight frameworks, lazy loading, and optimized images. Avoid unnecessary third-party scripts and keep things modular. In short, slim it down. Be conscious of the total size of each page and look for waste to eliminate bloat and unneeded file transfer. 
• Utilize page caching and Content Delivery Networks (CDNs): Content caching and CDNs  significantly reduce server load and network bandwidth by serving previously rendered and saved content vs. requesting a new rendering from the server on every page view.  
• Improve the user experience: A logical, easy to navigate website reduces the overall number of clicks and searches required to find content and reduces the overall server requirements and energy impact. 
• Choose a sustainable CMS: Headless systems or CMSs that support static generation (like Hugo or Next.js) can dramatically reduce server loads. A CMS such as Umbraco Heartcore not only provides this sort of responsible framework, the company as a whole is committed to sustainability in its product, its people, and its overall carbon footprint.  
• Green hosting: Use cloud providers committed to renewable energy (like Google Cloud or AWS with green commitments) or look into specialized green hosting platforms.

The Way Forward 

Digital transformation is unstoppable—but it doesn’t have to be unsustainable. With smarter design, conscious choices, and a shared sense of accountability, we can make the internet more efficient without sacrificing innovation. The future of tech isn’t just faster and smarter—it can be greener too.