Surging interest in artificial intelligence (AI) systems would add further strain to global electricity grids with the potential to rival the massive energy consumption of bitcoin. Thankfully, the premier cryptocurrency has shown us a way to mitigate the impact.
A doubling of data-center revenue at Nvidia Corp last quarter shows that demand for generative applications such as ChatGPT has not yet hit its peak. The US chipmaker is the key provider of shovels in this AI goldrush, but those processors are neither cheap nor lean. Its latest flagship, the GH200 Grace Hopper Superchip, which is the size of a postcard, draws up to 1,000 watts — equivalent to a portable heater.
Although most customers would be opting for something less fancy than the Superchip, they do buy them in bulk to connect together into a massive AI server and that is where the hunger for electricity really kicks in. One study published last year looked at the energy consumption required to train a single large-language model used to output text in multiple languages.
BLOOM from start-up HuggingFace drew on 176 billion parameters from 1.6 terabytes of data. It took a cluster of 384 Nvidia A100 graphics processing units (GPUs) more than 118 days to crunch, the study’s authors said. The electricity consumption from running so many GPUs for so long likely created 24.7 tonnes of carbon dioxide, they estimated. However, the true cost doubles to 50.5 tonnes when you take into account the network connections and idle time of the entire system.
Even then, training a model is just the start. According to one estimate from Amazon.com Inc., which runs its own AI servers, 90 percent of the expense from running artificial intelligence comes in the next phase when users query the model to get results — such as asking ChatGPT for chocolate-cake recipes.
The energy expenditure from implementing the data, called inferencing, is hard to calculate, but it is believed to be roughly in the order of 10 times that required in the first training phase — which means 500 tonnes of carbon dioxide. Moreover, a single generative AI query might have a carbon footprint four times larger than a Google search, one estimate showed.
Brute-force number crunching is built into bitcoin’s design, and helps explain why a wave of chips and servers was rolled out around the world in the hope of mining digital gold. An ongoing study at the University of Cambridge estimates that bitcoin is responsible for 72.5 million tonnes of carbon dioxide. That figure could be as low as 3 million tonnes if all bitcoin mines were run on hydroelectricity. (Researchers tend to use the term carbon dioxide equivalent, taking into account the emission of other greenhouse gases, which is then converted into an equivalent measure of carbon dioxide.)
Compared with the wastefulness of cryptocurrency, 500 tonnes of carbon dioxide from a single round of training and deployment seems like nothing. Yet it is still equal to driving 1.6 million kilometers in a gasoline-powered vehicle, or 500 flights from New York to Frankfurt.
And it is still early days. At least a dozen major technology companies are rushing to build and deploy generative AI products, including Amazon.com, Alphabet Inc, Microsoft Corp, OpenAI, Meta Platforms Inc, Baidu Inc, Tencent Holdings Ltd and Alibaba Group Holding Ltd. Since they are all in a race to outdo each other, they would not sit still once a model has been trained; they would keep buying power-hungry processors to analyze increasingly large amounts of data. Once that is done, they would compete with each other to serve up the results to consumers in the form of college essays, deepfake videos and synthetic versions of Pink Floyd music.
Adding insult to injury is that most AI training is powered by fossil fuels. These server farms have been expanded quickly in existing locations, often thousands of kilometers away from hydroelectric dams or solar power arrays. Since network latency is an issue when responding to Internet requests, they need to be near the end user and not located thousands of kilometers away.
However, bitcoin has already blazed a trail for the AI industry to follow. Cold climates with plenty of renewable energy became the perfect place to plonk down a power-hungry cryptomine, with the artic air and abundant thermal energy of Iceland making the country an ideal choice.
China might also find a new use for the many hydroelectric power stations that attracted mining rigs, but that lost business after it cracked down on the digital currency. Foreign AI providers would not be able to tap in, naturally, but Chinese technology giants know they have such a resource close by as their power needs rise.
There is another upside to switching out bitcoin for AI in these server farms. While the cryptocurrency attracted numerous speculators and billions of dollars in investment, it still failed to add much value to the world. Generative artificial intelligence does not have that problem; just ask ChatGPT.
Tim Culpan is a Bloomberg Opinion columnist covering technology in Asia. Previously, he was a technology reporter for Bloomberg News. This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.
In the US’ National Security Strategy (NSS) report released last month, US President Donald Trump offered his interpretation of the Monroe Doctrine. The “Trump Corollary,” presented on page 15, is a distinctly aggressive rebranding of the more than 200-year-old foreign policy position. Beyond reasserting the sovereignty of the western hemisphere against foreign intervention, the document centers on energy and strategic assets, and attempts to redraw the map of the geopolitical landscape more broadly. It is clear that Trump no longer sees the western hemisphere as a peaceful backyard, but rather as the frontier of a new Cold War. In particular,
As the Chinese People’s Liberation Army (PLA) races toward its 2027 modernization goals, most analysts fixate on ship counts, missile ranges and artificial intelligence. Those metrics matter — but they obscure a deeper vulnerability. The true future of the PLA, and by extension Taiwan’s security, might hinge less on hardware than on whether the Chinese Communist Party (CCP) can preserve ideological loyalty inside its own armed forces. Iran’s 1979 revolution demonstrated how even a technologically advanced military can collapse when the social environment surrounding it shifts. That lesson has renewed relevance as fresh unrest shakes Iran today — and it should
The last foreign delegation Nicolas Maduro met before he went to bed Friday night (January 2) was led by China’s top Latin America diplomat. “I had a pleasant meeting with Qiu Xiaoqi (邱小琪), Special Envoy of President Xi Jinping (習近平),” Venezuela’s soon-to-be ex-president tweeted on Telegram, “and we reaffirmed our commitment to the strategic relationship that is progressing and strengthening in various areas for building a multipolar world of development and peace.” Judging by how minutely the Central Intelligence Agency was monitoring Maduro’s every move on Friday, President Trump himself was certainly aware of Maduro’s felicitations to his Chinese guest. Just
On today’s page, Masahiro Matsumura, a professor of international politics and national security at St Andrew’s University in Osaka, questions the viability and advisability of the government’s proposed “T-Dome” missile defense system. Matsumura writes that Taiwan’s military budget would be better allocated elsewhere, and cautions against the temptation to allow politics to trump strategic sense. What he does not do is question whether Taiwan needs to increase its defense capabilities. “Given the accelerating pace of Beijing’s military buildup and political coercion ... [Taiwan] cannot afford inaction,” he writes. A rational, robust debate over the specifics, not the scale or the necessity,
RSS