An incident on 28 March 1979 changed the course of energy forever, at least in the US. The Three Mile Island Nuclear Generating Station at Pennsylvania had a partial meltdown of a reactor core. No injuries were reported, but the reactor was mothballed, and it heightened public fears of nuclear power.
So, it came as a bolt from the blue when Three Mile owner Constellation said last week that it will restart the nuclear reactor and Microsoft will buy the energy generated for 20 years.
The reason: to meet its burgeoning need for clean energy for its massive artificial intelligence (AI) data centres, as the tech industry balances building giant energy-sucking AI models with its environmental, social and governance goals.
The GenAI revolution sparked off by the launch of ChatGPT in 2022 has catalysed an insatiable appetite among tech firms and countries to build bigger AI models in their race towards artificial general intelligence (AGI) and global domination.
AI bots are adding hundreds of millions of users and promise to reshape industries and societies. As Big Tech CEOs commit hundreds of billions of dollars to lead this era-defining technology, they also face huge obstacles in scaling up.
Data availability is one, the price of Nvidia graphics processing units (GPUs) is another, but the biggest perhaps is the availability of clean and reliable power for both the training and inferencing needs of large language models.
For example, Microsoft and OpenAI are contemplating a $100 billion AI supercomputer, and the five gigawatts it would take to power it is what’s required for all of New York City today!
The Electric Power Institute expects data centres to consume up to 9% of US electricity generation by 2030, more than double what they currently use. This translates to 50 gigawatts (GW) of additional power, a colossal amount of energy which could power 10 New Yorks.
Moreover, it requires an investment of over $500 billion in data centre infrastructure alone. Most of the power generated globally is from coal and other fossil fuels, making it a top contributor to carbon-dioxide emissions and global warming.
The global concern this spawned has led Big Tech firms to declare aggressive net-zero targets, which would mean they must rely on green power for almost all their needs.
They have signed long-term power purchase agreements (PPAs) with wind and solar power companies, but as the Electric Power Institute reports, those deals “typically do not match electricity demand hour by hour with local resources” and so there is “no guarantee that all electricity-related greenhouse gas emissions are offset.”
Internal data released by Microsoft, Google and others supports this claim. Thus, the gold rush for clean power. Other than the Three Mile Island revival, Microsoft has made other moves.
It has tied up with Brookfield Asset Management in a $10 billion deal to develop another 10.4GW of renewable energy capacity across the US and Europe.
It has even started dabbling in nuclear fusion energy, with an agreement with Helion Energy and by hiring people to build small modular nuclear reactors. Google is signing PPAs with companies to procure 100 megawatts of offshore wind farm energy.
The question is whether it is worth chasing the AI dream, with all this potential damage and what it will cost to undo it through clean power generation. The tech industry thinks so. McKinsey estimates (bit.ly/47DYYB2) that GenAI could help create $2.6-4.4 trillion in economic value for the global economy.
Sam Altman believes Artificial Super Intelligence will help humanity solve big problems like climate change, longevity and nuclear fusion, thus making the investments worthwhile. Even Bill Gates has said, “The question is, will AI accelerate a more than 6% reduction [in emissions]? And the answer is: certainly.”
He also believes that Big Tech businesses would pay a ‘green premium’ for clean energy, thus incentivising its development and deployment over fossil fuels, and so the AI rush would be worth it.
Every technology has raised such existential questions, and none of them has been as consequential as AI promises to be. Another such technology was nuclear, but the spectre of Hiroshima and Nagasaki made a fearful world clamp down on it.
While there has not been another nuclear attack since 1945, the flip side was that we slowed down nuclear power generation in favour of coal and gas, which, in turn, worsened global warming. There were no easy solutions then and there aren’t any now.
#artificial #intelligence #worth #money #energy #guzzling