Google's Emissions Jumped 50% in 2024 to Feed Generative AI

Google reported a 50% increase in its greenhouse gas emissions in 2024, driven by the compute demands of generative AI — one of the clearest signals yet that the AI buildout has broken the climate commitments Amazon, Microsoft, Google and others made in the 2010s.

The figure anchors a broader accounting laid out in "Hyperscale: The Ambition and Excess of Big Tech's Data Empires," a new book by Paris Marx, host of the Tech Won't Save Us podcast. Marx visited data centers from Ashburn, Virginia, to South Dublin, Ireland, and documented a physical infrastructure of computation that now rivals heavy industry in its demand for electricity, water and fuel.

The stakes are structural. Amazon, Microsoft and Google alone control 59% of hyperscale data center capacity, according to the book. Every major AI product — from OpenAI to Meta's models — runs on this infrastructure. The question the book forces is whether the emissions, water consumption and local health costs of that infrastructure are justified by what AI actually delivers.

How much power does a hyperscale data center need?

The numbers have shifted by an order of magnitude in a decade. A large data center once required around 50 megawatts of power. Hyperscale facilities can require triple that. Meta's planned Louisiana data center may require 4,500 megawatts — 4.5 gigawatts.

Officials say that figure is four times more than all the electricity used by New Orleans on its hottest day in 2025. It exceeds the two gigawatts produced by the Hoover Dam.

Water demand has scaled alongside power. AI-era chips such as GPUs run hotter than their predecessors, which pushes facilities to add water-based cooling on top of air conditioning. A hyperscale center can consume 1 million to 5 million gallons of water per day — the daily usage of a town of more than 10,000 residents, according to a 2025 estimate cited in the book.

Marx frames the security surrounding these facilities as a measure of how seriously the industry guards them. He described tall fences, pervasive security cameras, badge-controlled doors and, outside a cluster of Microsoft-owned centers in Ireland, groundskeepers mowing grass that did not need cutting.

"It was just really staggering to me that it was at that level that it was so hard to move through these spaces," Marx told CNET.

Why are data centers burning fossil fuels?

Grid connection is the bottleneck. The approval and interconnection process for a new data center can take years, so tech companies have started building their own power plants next to the facilities for the sole purpose of keeping them running.

Those plants burn fossil fuels — gas turbines, on-site diesel tanks held in reserve for outages, or coal. That choice has direct consequences. Communities neighboring data centers have reported brown or unsafe drinking water and constant noise. A lawsuit against Elon Musk's xAI centers in Memphis cites extreme health risks.

Memphis residents "are feeling it in their lungs because of the fossil fuels that are being burned," Marx said.

The local effects compound a global one. Decades of climate research link additional fossil-fuel combustion to worsening extreme weather, health risks and natural disasters. Marx writes in the book: "The world will suffer through the worsening natural disasters fueled by the quest for computation by some of the most powerful people on the planet."

What happened to the carbon-neutral pledges?

In the 2010s, Amazon, Microsoft, Google and many other tech companies pledged some version of carbon neutrality — or carbon negativity — within the following decades, committing to offset their emissions with at least equivalent emission-reducing actions.

Sustainability reports from that period show the companies were measurably shrinking their environmental footprints. The trend then reversed.

"And then ChatGPT comes along, and all of a sudden, that's out the window," Marx told CNET.

Industry executives have acknowledged the shift. Microsoft President Brad Smith told Bloomberg in 2023 that the company's pledge to be carbon negative by 2030 "was before the explosion in artificial intelligence," and that the carbon goal was now "five times as far away as it was in 2020" because of AI. Google's 50% emissions increase in 2024, disclosed in its own reporting, marks the same trajectory.

The reversal matters beyond corporate scorekeeping. The hyperscale operators making these pledges are also the dominant cloud and AI providers, and their combined capacity share means their energy decisions move regional grids and local environments at scale.

Is the trade worth it?

Anti-data-center movements in Memphis, Oregon, Chile, Ireland and thousands of other communities have answered no, and Marx's book positions that opposition as one of several fronts on which communities are pushing back against Big Tech's power.

Marshaling the resource case against the buildout, Marx asks: "How much computation do we really need here? How many resources are we willing to dedicate to this vision of the future? And is this technology really delivering in such a way that expending all those resources is justified?"

The answer — for many data center neighbors across the globe — is, in the book's words, an unflinching no.

Unless the industry decouples AI compute from on-site fossil generation, every capability gain in models translates directly into fuel burned, water drawn and emissions booked against pledges the companies once presented as binding.