How Big Tech’s AI Competition is Harming Our Planet
By Richard Anthony, Emerging Technologies Policy Advocate; Asha Buerk, Summer Technology Policy Intern
The exponential rise of artificial intelligence (AI) has big tech corporations scrambling to overtake each other. But as megacorporations rush to be at the cutting edge of new AI models, the implementation of green energy and carbon neutrality policies have seemingly fallen by the wayside. For instance, Google’s recent feature, AI summaries, cause every search to require over 30 times more energy than a typical query, while searching via ChatGPT uses 10 times as much energy. A generic natural language processing model emits over 300,000 kgs of CO2 while being trained, which doesn’t even begin to cover usage emissions. By 2040, emissions from internet communication technologies will account for over 14% of global outputs and AI specifically will be responsible for 21% of all energy usage. The massive amounts of energy used to train, modify, and use these models is often ignored or brushed aside by major organizations amidst the chaos of the AI race. Their outsized gobbling of energy demand is upending domestic power markets, risking higher prices for working families’ utility bills.
Yet, even though big tech companies regard AI as the future of the market, other organizations are not quite as sure. A recent report released by Goldman Sachs claims that AI in its current state, and even as far as 10 years down the line, does not deliver on its economic promises. We’re seeing a rapid and catastrophic amount of energy misuse and carbon release for a product whose consequential output or meaningful human assistance capacity may be more than a decade away, at the earliest.
Even so, tech companies continue on in the AI race, and consequently have backslid on their promises or goals of carbon neutrality. Microsoft, for one, is 30% over its expected emissions output, which it cites is directly due to its advances in the AI industry. Similarly, Google is over 50% over its intended emission amount due to its recent foray into, which does not bode well for its promise to be carbon neutral by 2030, especially with a new billion dollar data center to power their models on the way.
In order to create and maintain top-of-the-line models at a breakneck pace, companies have chosen to keep fossil fuel-powered electricity centers open, rather than closing them and moving to solar and green energy sources, as originally intended. The AI race is so intense that AI companies like Meta say they believe that energy demand will constrain their ability to build out as much data center capacity as they would like. Any promises or intentions of ‘going green’ are fundamentally incompatible with the current mindset big tech companies have with the AI market.
In the United States, these data centers are heavily concentrated in just a few areas, concentrating pollution and related harms. a. These fossil fuel-powered facilities are primarily located in Northern Virginia, Arizona, Ohio and Utah, where residents have cited noise complaints, dangerous high-voltage lines through or near their properties, and high pollution rates that make it difficult to live alongside the facilities. Data centers turn quiet, rural areas into whirring centers that bring a handful of jobs, but inevitably displace residents due to their loud and disruptive power lines and facilities. As these markets and centers get the last bits of power “squeezed out” of them, tech companies are scrambling to find other locations to house their data centers, worsening living conditions for new swaths of individuals.
To “combat” these high emission rates, big tech companies have taken to purchasing large amounts of carbon offsets in an attempt to satisfy their carbon-neutral promises. But this strategy does nothing to actually make their processes more sustainable or eco-friendly. Additionally, carbon offsets have repeatedly been shown to have dubious value, including because they often involve carbon savings that were going to be achieved anyway.
Several options exist for companies looking to mitigate and reduce the nonrenewable energy needed for AI systems. For one, corporations could begin to reduce or slow their rollout of commercial generative AI systems. This change would allow consumers access to the same products that they’re enjoying, but would allow companies to continue their switch to green-energy-powered data centers and begin to use carbon offsets for their explicit purpose: alongside carbon free pursuits. Additionally, setting caps on the amount of power GPU’s intake during the training and deployment phases can reduce its carbon emissions by up to 15%. Changing the way AI models are trained can massively reduce energy demand during training. Establishing federal energy efficiency standards for data centers is another long-overdue reform. Technical solutions like these exist and are viable for companies to pursue, but require effort and without encouragement via legislation, will likely not be pursued.
Amidst our current climate crisis, big tech’s disregard toward the environmental impacts of AI models is irresponsible and deeply harmful towards both local communities and global emissions. Policy makers must insist companies to continue with their plans of switching to eco-powered data centers to sustain consumer experience, rather than backsliding for corporate profit in the AI space.