Hacker News
Carbon Emissions of US Data Centers Rival Countries
Kim_Bruning
|next
[-]
nisegami
|next
|previous
[-]
filoeleven
|root
|parent
[-]
> Natural gas and coal together are expected to meet over 40% of the additional electricity demand from data centres until 2030.
https://www.iea.org/reports/energy-and-ai/energy-supply-for-...
nisegami
|root
|parent
[-]
vivzkestrel
|next
|previous
[-]
- https://medium.com/@fareedkhandev/a5792cbf3b59?sk=7772001b26...
- if this is indeed true, wouldnt it render all data centers obsolete?
dan_grec
|next
|previous
[-]
The US is once again making piles of cash for some people, but at what cost?
rayiner
|root
|parent
[-]
AI companies are losing money hand over fist. It's a highly competitive market and right now the benefits are going to the consumer, not the provider. The environmental costs should be attributed to the ultimate consumer who gets the benefit, not the producer that is, for now, losing money providing the service.
Most of the industries that generate the most carbon emissions are competitive industries where price competition decimates profits and causes most of the surplus to flow to the consumer. Exxon's profit margin over the last 5 years averages 10%. Even right now with the high gas prices, their profit margin for the last quarter was only 14%. These companies have enormous revenues, but they also spend huge amounts of money turning dead dinosaurs into usable fuel. Most of the benefit goes to American consumers, so they can live spread far apart from each other in houses twice as big as those in Europe and drive their fat asses through the drive through to get their morning Starbucks.
By all accounts, AI is going to be a continuation of this trend. We'll have multiple competing companies driving prices to efficient levels, and most of the surplus will go to consumers so they can turn their brains off while drinking their Starbucks and watching Tik Tok in the background.
hollerith
|root
|parent
[-]
Since 2021, OpenAI has raised about $192 billion and Anthropic has raised $245.9 billion including today's IPO. Both are spending a large fraction of that money in an attempt to create models that are much more powerful than the models they have today. I'm tempted to suggest that the main reason they offer services to the public and generate revenue is to make themselves more attractive to investors. (E.g., they hope that prospective investors will interact with Claude or ChatGPT and be impressed with the result that they choose to invest in the company.)
In July 2025, WIRED reported that Meta had offered some researchers compensation packages worth up to $300 million over 4 years, including equity. Meta was willing to offer that much because like OpenAI and Anthropic, they are trying for a model that is much more powerful than any of today's models. It wouldn't make sense to offer that much to a single researcher in the hopes that that researcher will help Meta make the process of serving inference more efficient, which you seem to think will be the main arena of competition between the labs.
(I've focused on OpenAI and Anthropic because Meta and Google are harder for an outsider to analyze because they have diverse businesses and are required only to report company-wide financials, not the financials of every specific business they are in.)
The nation's largest grocery retailers don't spend many 10s of $billions trying to create grocery stores or retail experiences that are vastly better than the stores or experiences they offer today.
The world's petroleum producers do spend vast sums every year to get petroleum, but they don't spend $10 of billions in addition to that in an attempt to get a product that is qualitatively superior to the product they currently offer.