Comments by OSFR historian Jim Tatum.
jim.tatum@oursantaferiver.org
– A river is like a life: once taken,
it cannot be brought back © Jim Tatum
In 2011, when the U.S. economy was still dazed by the Great Recession shockwave, Marc Andreessen, an influential venture capitalist and tech luminary, wrote an opinion piece in the Wall Street Journal.
Andreessen, who founded the Netscape web browser in the early days of the public internet, argued that the leading U.S. tech companies – industry veterans like Microsoft and Apple, plus comparative newcomers such as Amazon and Google – were the vanguard of an economic revolution. Their digital products were toppling the business models of the analog era. “Software,” he declared, “is eating the world.”
Andreessen hit the mark. The tech revolution he identified has remade economies and societies. In recent years Amazon, Apple, Google, Microsoft and their peers have continued churning, gathering speed, and expanding in new directions. Software did indeed eat the world. And now, with data centers rising like mushrooms after a storm, it’s hardware’s turn at the table.
The AI revolution is to blame. Massive amounts of computing power are needed to train and run AI models. Boosters proclaim world-changing outcomes. Detractors find it difficult to wade through the slop. Boom or bust, the investments and resources dedicated to AI today have profound implications not only for workers and economic growth. The magnitude of this change will influence the future of the world’s energy use – and by extension, its water and its climate.
Looking back, 2025 is the year that the pace and scale of the AI boom and its natural resource implications came into sharp focus for the public and policymakers.
This year was a reminder that the virtual worlds that lie behind our computer and cell phone screens have a direct foothold in the physical world.
It is hard to understate the heights of the AI transformation happening now. In investment terms, the closest comparison might be the railroad mania of the late 19th century. The $500 billion Stargate project in the U.S. is one piece of what is projected to be a multitrillion-dollar global buildout of computing infrastructure in the next few years. In energy terms, AI is a voracious beast. The data centers at the heart of it, especially those that train new AI models, require prodigious loads of electricity. Industry observers again grasp at historical comparisons that reach back generations.
“We’re really talking about load growth that we have not seen in this country since we tried to electrify rural America and put a refrigerator and washing machine in everybody’s house,” said Abe Silverman, an energy researcher at Johns Hopkins University, referring to the post-World War Two era. “The speed at which these new loads are coming on is so intense, and the political interest and economic development and AI expansion is so intense that we’re really running into a challenge that the system can’t expand fast enough to meet the new load.”
In a December 2024 report, Lawrence Berkeley National Laboratory estimated that data centers consumed about 4.4 percent of U.S. electricity. That number could rise, the report said, to 6.7 percent and 12 percent by 2028. A year later, energy experts say that the trend line is probably tracking the higher end of that range.
A closer look at the numbers reveals why. The electricity demand for a single new AI-training data center is 1 gigawatt or more. That’s more electricity than the entire state of Wyoming consumes. Hundreds of such facilities are in various stages of development.
The surge is most vigorous in the U.S. But it is also a global phenomenon. In October, Google announced a $15 billion data center hub in southern India, the company’s largest outside its home country. One consultancy expects data center capacity in the six Middle Eastern countries of the Gulf Cooperation Council to triple by 2030. China is the world’s second-largest source of data center electricity demand, and it is projected to rise at a faster rate than in the U.S.
We’re really talking about load growth that we have not seen in this country since we tried to electrify rural America and put a refrigerator and washing machine in everybody’s house.
Abe silverman, johns hopkins university
The energy demand in this computing revolution is important for many reasons, but especially for the world’s water and climate.
Data center equipment generates heat that must be removed. Often water is used as a coolant, and this has driven up local water use. Zero-water or low-water options exist, but they increase energy use, which has both carbon and water implications. The electricity that powers the data centers, if it comes from sources other than wind and solar, consumes water when it is generated. The water embedded in energy use for data centers is a much larger number than the water used in direct cooling operations. Exact numbers are difficult to acquire; companies tend to guard facility-level data.
These inputs are more than an accounting exercise. Already, the world is off track on its climate goals. Carbon emissions in 2025 are expected to reach a record high, according to Global Carbon Budget, a research group. Utilities like Arizona Public Service have announced new natural gas power plants to meet growing data center energy demand. Data centers in 2024 accounted for only 0.5 percent of global carbon emissions, according to the International Energy Agency. But it is one of the few sectors that is growing rapidly. By 2030, in a mid-range scenario, data centers account for 1 percent of global emissions.
A carbon-intensive data center buildout would make a risky global climate even riskier. Research published earlier this year found that multi-year droughts globally are “increasing in occurrence, frequency, and severity.” In Iran, the worst drought in decades has exposed dismal failures in its water management approach. The country is in a full-on crisis. Its reservoirs are on the brink of drying up and its president is talking about moving the capital from Tehran. Warming oceans, meanwhile, have fueled the rapid intensification of tropical storms, which are battering vulnerable coastal regions, especially Southeast Asia this year.
As data center hardware is eating the world, public resistance is growing. The City Council in Tucson, Arizona, citing residents’ water concerns, rejected a proposed data center in August. In Delaware, the New Castle County Council is debating data center regulations. Minnesota passed a law in June that applies stricter water permitting for data centers using more than 100 million gallons annually. Georgia strengthened its public review process.
Which is not to say that AI and its derivatives are only to be feared. Already these tools are proving to be valuable. AI models are fine-tuning weather and river-flow forecasts. Google’s Flood Hub is giving officials earlier warnings of flood risk. Add sensor data, and water utilities are using high-end computing power to identify leaks and optimize capacity in their sewage systems in order to reduce spills and cut costs.
The other bit of good news is that renewable energy sources are also experiencing bumper times. Globally, electricity generation from renewable energy – including solar, wind, hydro, geothermal, and biomass – exceeded that from coal for the first time ever in the first half of 2025. In the U.S., the federal government’s energy researchers say that solar photovoltaic is in many cases the cheapest source of new electricity.
These energy, water, and carbon calculations for data centers represent an intrusion of reality in an industry whose visions of the future can seem unbounded. This year was a reminder that the virtual worlds that lie behind our computer and cell phone screens have a direct foothold in the physical world. If software and now hardware are eating the world, it is not only obsolete business models that are part of the buffet. It is also the stuff of life: minerals and metals, land and water.
*image: Wikipedia, Creative Commons
