So much for green Google ... Emissions up 48% since 2019

With six years left for Google to meet its 2030 "net zero" climate commitment, the web giant has admitted its carbon emissions are rising.

In an environmental report [PDF] published on Tuesday, Google admitted that its greenhouse gas emissions rose 48 percent since 2019. In 2023 alone, the search giant's carbon footprint topped 14.3 million tons of CO₂ equivalents – up 13 percent from the prior year.

Chief sustainability officer Kate Brandt and SVP of learning and sustainability Benedict Gomes pinned the blame squarely on increased datacenter energy consumption.

This shouldn't come as much of a surprise, considering Google operates enormous datacenter fleets, and the sheer amount of compute required to train and deploy its growing library of AI models and services. Training the largest models often requires tens of thousands of accelerators running flat out for weeks or even months.

Everyday Google services like web search also consume more power these days, thanks to machine learning: It's estimated that an AI-powered query consumes ten times more energy than a standard search.

AI's impact on the environment wasn't lost on Brandt or Gomes, who spun it as a net positive. "We know that scaling AI and using it to accelerate climate action is just as crucial as addressing the environmental impact associated with it," Brandt and Gomes wrote in the report.

Google's argument is that AI development contributes to a larger carbon footprint, but the technology will offset more emissions elsewhere – such as by generating more efficient routes for automobiles and other ways to cut emissions.

Google estimates that AI technologies could mitigate five to ten percent of global greenhouse gas emissions by 2030. Of course, the Chocolate Factory also admits that there's still a lot of "uncertainty" surrounding the long-term impact of AI on the environment, making it "complex and difficult to predict."

Google remains committed to its 2030 net-zero target, but the report admits things could get worse before they get better.

"As we further integrate AI into our products, reducing emissions may be challenging due to increasing energy demands from the greater intensity of AI compute, and the emissions associated with the expected increases in our technical infrastructure investment," the report reads.

Google isn't the only cloud provider grappling with a growing carbon footprint. In May, Microsoft revealed its carbon dioxide emissions had increased nearly 29.1 percent since 2020.

Redmond, which provides a large quantity of compute to OpenAI, also blames the increase in emissions on the construction and provisioning of datacenter facilities for its cloud and AI business units.

• Bill Gates says not to worry about AI gobbling up energy, tech will adapt
• AI to boost datacenter capex by 28.5% and become the top server workload
• Etched looks to challenge Nvidia with an ASIC purpose-built for transformer models
• Want to save the planet from AI? Chuck in an FPGA and ditch the matrix

AI's appetite for power has become a source of concern. Most industry experts concur that AI will have a material effect on datacenter energy consumption going forward, but few agree on the extent of that impact.

In January, the International Energy Agency estimated that global datacenter power consumption could double by 2026.

Meanwhile, a Goldman Sachs analysis [PDF] from April 2024 predicted that global datacenter power use would increase by between 1.8x and 3.4x by the end of the decade.

Much of this AI growth is expected to take place in the US, according to a Rystad Energy report from last month that forecast datacenter power consumption to double by 2030.

However, others have disputed such predictions. In February, the Information Technology and Innovation Foundation's Center for Data Innovation (CDI) – a US think tank backed by many of the largest tech corporations – made the case that AI's long-term emissions impact is being overblown.

The group argued that past attempts to accurately predict the energy demand of emerging technologies have failed, and that many of these forecasts don't consider improvements to hardware and software. ®