Why Are Data Centers and AI Growing So Fast? A Skeptic’s Guide to What Is Actually Driving the Boom
Why Are Data Centers and AI Growing So Fast? A Skeptic’s Guide to What Is Actually Driving the Boom
- Data centers were expanding long before generative AI because cloud computing, streaming, business software, online services, and digital storage kept growing.
- AI accelerates that trend because modern models use specialized servers that can require much more computing power than conventional workloads.
- The International Energy Agency reported that electricity consumption from AI-focused data centers increased about 50% in 2025.
- The boom is real, but its final size remains uncertain because efficiency improvements, grid bottlenecks, financing, and AI adoption could slow expansion.
- A skeptical view does not require dismissing the entire trend. The useful question is which part reflects durable computing demand and which part reflects optimistic investment expectations.
If someone is skeptical about the explosion of data centers and artificial intelligence, telling them that “AI is the future” probably will not accomplish much. It sounds suspiciously similar to every technology pitch humans have produced immediately before spending heroic amounts of money.
A better explanation starts with something less dramatic: the data-center boom did not begin with ChatGPT or generative AI. For years, businesses and consumers have moved computing, storage, software, entertainment, communications, and financial activity into large centralized facilities. AI is now adding a particularly computation-intensive layer on top of that existing trend.
The growth is therefore neither entirely hype nor perfectly predictable. There is real demand underneath it, but there is also uncertainty about how much infrastructure the market will ultimately need.
1. Data Centers Were Growing Before the Current AI Boom
The first thing to explain to a skeptic is that data centers are not synonymous with AI. They already form much of the physical infrastructure behind the modern internet and cloud economy.
When someone streams a movie, stores photos in the cloud, pays a bill online, joins a video meeting, searches the web, plays an online game, or uses business software hosted remotely, servers somewhere are performing the work. The same is true for countless less visible corporate systems involving databases, logistics, cybersecurity, communications, and analytics.
That computing used to be distributed more heavily across company-owned server rooms and local hardware. Over time, much of it shifted toward large cloud and colocation facilities because centralized infrastructure can be easier to scale, maintain, secure, and utilize efficiently.
Lawrence Berkeley National Laboratory found that conventional server electricity use in the United States also grew significantly during the past decade. In other words, even without the latest AI models, digital demand was already moving upward.
This distinction matters because it prevents a common misunderstanding. If generative AI growth slows tomorrow, data centers do not suddenly become unnecessary. The cloud economy underneath them would still exist.
2. AI Changes the Equation Because It Requires Much More Specialized Computing
AI growth matters because many modern AI workloads rely on GPU-accelerated servers designed to perform enormous numbers of calculations in parallel. That pushes computing density and electricity demand higher.
Traditional web applications may store information, process transactions, or serve pages to users. Large AI systems perform different kinds of computational work. Training can involve processing enormous datasets across large groups of specialized chips, while inference means running trained models each time users or software request an AI-generated result.
The 2024 Lawrence Berkeley National Laboratory report estimated that U.S. GPU-accelerated AI server electricity consumption rose from less than 2 TWh in 2017 to more than 40 TWh in 2023. That does not represent all data-center electricity use, but it illustrates how quickly accelerated computing became significant.
The trend continued. The International Energy Agency reported that global data-center electricity consumption increased about 17% in 2025, while electricity consumption from AI-focused data centers increased roughly 50%.
That is the physical reason behind much of the construction boom. Companies are not merely storing more files. They are installing a new class of computing equipment that can require much more electricity, cooling, networking equipment, and supporting infrastructure.
3. Why Are Companies Building So Much Capacity Before AI Demand Is Fully Proven?
Data centers take years to plan, connect to the grid, finance, equip, and build. Companies therefore have to invest before they know exactly how large future AI demand will become.
This is one reason the current expansion can look irrational from the outside. Technology changes quickly, but power infrastructure does not. A software company can release a new AI model in months. A transmission project, substation, power plant, or major utility interconnection can take much longer.
The IEA notes that a data center itself can sometimes become operational within two or three years, while the broader energy system typically has longer planning and construction timelines. Companies expecting future computing demand therefore compete for land, electricity connections, chips, transformers, cooling equipment, and construction capacity well in advance.
There is also a competitive reason to build early. If AI becomes deeply integrated into search, office software, programming, advertising, science, customer service, logistics, entertainment, and other industries, companies without enough computing capacity could find themselves unable to meet demand. Building too little can therefore be strategically expensive.
Building too much is obviously expensive too. That is precisely why a skeptical interpretation remains legitimate. The industry is effectively placing very large bets on future AI use before anyone can know exactly how profitable all those applications will become.
4. Is the Growth Really as Large as the Headlines Suggest?
The growth is large, especially in the United States, but global context matters. Data centers are becoming an important new electricity load without consuming anything close to a majority of the world's electricity.
The U.S. Department of Energy reported that data centers consumed roughly 176 TWh of electricity in 2023, representing about 4.4% of total U.S. electricity use. Lawrence Berkeley National Laboratory projects a wide 2028 range of approximately 325 to 580 TWh, or about 6.7% to 12% of U.S. electricity consumption.
Globally, the IEA's updated central projection sees data-center electricity consumption increasing from about 485 TWh in 2025 to approximately 950 TWh in 2030. That would represent around 3% of global electricity demand.
Both things can therefore be true at once. Data centers can create serious electricity-system challenges in particular regions while remaining a relatively modest share of global electricity consumption. The facilities are unusually concentrated geographically, which is why their local effect can be much larger than the global percentage suggests.
That concentration explains why utilities, regulators, and communities can experience the boom very differently. A global 3% figure sounds manageable. Several enormous projects attempting to connect to the same regional grid at roughly the same time can be considerably less charming.
5. What Should a Skeptic Be Skeptical About?
Skepticism is most useful when it focuses on uncertain assumptions: future AI adoption, profitability, efficiency, financing, grid availability, and whether today's projected facilities will actually be built.
The first uncertainty is demand. AI usage has expanded rapidly, but the industry still does not know which applications will become everyday necessities and which will remain expensive demonstrations looking for a business model. Today's infrastructure forecasts depend heavily on future adoption.
The second uncertainty is efficiency. AI hardware, software, and model design continue improving. The IEA reports that energy use per AI task is declining rapidly. If future systems perform the same work with dramatically less electricity, the industry may need less power than today's most aggressive forecasts assume.
But efficiency can work in the opposite direction at the system level. Cheaper AI can encourage far more usage, while newer applications such as video generation, advanced reasoning, and AI agents can require substantially more computation than simple text requests. Falling energy use per task therefore does not necessarily mean falling total electricity consumption.
Finally, there are physical and financial constraints. The IEA notes tightening supplies of transformers, turbines, advanced chips, and other equipment, along with grid-connection and permitting delays. It also warns that the pace of investment depends on market expectations and access to financing. A proposed project is not the same thing as an operating data center.
Key Takeaways at a Glance
- Data-center growth is broader than AI. Cloud services and conventional digital workloads were expanding before the generative-AI boom.
- AI substantially increases computing intensity. GPU-accelerated servers are a major new source of electricity demand.
- Companies are investing ahead of demand. Infrastructure takes years to build, so current construction reflects expectations about future AI adoption.
- The growth is significant but uncertain. Efficiency, financing, supply chains, and grid bottlenecks could change the trajectory considerably.
- Skepticism and recognition of real demand can coexist. The central question is how much of today's investment will produce durable economic value.
| Growth Driver | Why It Is Real | Reason for Skepticism |
|---|---|---|
| Cloud computing | Long-running digital migration | Efficiency limits demand growth |
| AI training | Requires large specialized computing clusters | Future model economics are uncertain |
| AI inference | More users create recurring workloads | Energy per task keeps improving |
| Infrastructure investment | Companies need capacity before demand arrives | Some projects may be delayed or canceled |
| Electricity demand | Measured consumption is already rising | Long-term forecasts remain wide |
The Most Reasonable View Is Somewhere Between Hype and Dismissal
The simplest explanation for the data-center boom is that society is asking computers to do more things, and AI asks them to perform some particularly demanding things. More digital services create more server demand, while AI adds specialized high-performance computing on top of that foundation.
That does not prove every investment forecast is correct. Technology industries routinely overshoot, discover bottlenecks, improve efficiency, consolidate, and occasionally build expensive monuments to assumptions that seemed brilliant three years earlier.
But dismissing the entire expansion as an AI bubble also ignores measured changes already occurring in server electricity consumption, cloud demand, GPU deployment, and grid planning. There is genuine infrastructure demand underneath the speculation.
For a skeptic, the useful approach is not to decide whether AI is universally transformative or universally overhyped. Watch actual usage, revenue, computing efficiency, electricity consumption, project completion rates, and whether businesses continue paying for AI services. Those indicators will reveal which parts of the boom survive once enthusiasm has to coexist with economics.
Sources
International Energy Agency • Key Questions on Energy and AI — Executive Summary
International Energy Agency • Energy Demand from AI
Lawrence Berkeley National Laboratory • 2024 United States Data Center Energy Usage Report
