This article is from the WeChat public account:Giant Tide WAVE (ID: WAVE-BIZ), Author: Xie Zefeng, Editor: Yang Xuran, Cover image from: Visual China
“The end of AI is photovoltaic and energy storage! If we only consider computing, we would need to burn the energy of 14 Earths; super AI will become a bottomless pit for electricity demand.”
Recently, this statement from AI guru Jensen Huang has gone viral. In fact, this is not Huang’s original wording; his original intent was that Nvidia helps improve computational efficiency and reduce energy consumption, and if computing speed does not accelerate, “we might need 14 different planets, 3 different galaxies, and 4 suns to fuel all of this.”
It is evident that artificial intelligence has an urgent demand for energy. OpenAI founder Sam Altman has also expressed a similar viewpoint; he once judged:
The future of artificial intelligence requires breakthroughs in energy because the electricity consumed by artificial intelligence will far exceed people’s expectations. In particular, nuclear fusion or cheaper solar energy and storage are the future directions of AI development.
Elon Musk has expressed more clearly that after the chip shortage comes the electricity shortage, “next year the electricity will not meet all chip demands.”
The high energy consumption of AI is already an undeniable fact; data centers and supercomputing centers are veritable “electricity monsters.” The United States’ power infrastructure has long been outdated, and although Europe has invested in new energy, it is still insufficient.
Looking globally, the second industrial revolution driven by electricity has not yet fully spread; many countries and regions still cannot obtain stable electricity supply, let alone consume and use electrical appliances.
At the same time, the so-called photovoltaic, lithium battery, wind energy, and other new energy sectors all require significant traditional energy support. Without coal, oil, and natural gas as a foundation, various so-called Nth industrial revolutions are like water without a source, trees without roots.
The end of AI ultimately returns to power generation and energy; this is no longer a joke, but a fact.
1. Electricity-Eating Beasts
The tension in computing power is behind the surge in energy demand.
Recently, the AI assistant Kimi developed by the Dark Side of the Moon has exploded in popularity. It supports uploading various documents and summarizes key content; the lossless context length it can handle has increased from 200,000 Chinese characters to 2 million characters. This means Kimi can read the entire book of “Zhen Huan Zhuan” in one go and can answer detailed questions about the drama, comparable to a “Zhen” scholar who has watched the drama dozens of times.
Due to the continuous surge in access volume, the Kimi application terminal once crashed, prompting the Dark Side of the Moon to conduct five expansion works.
The ability to handle super long texts and the continuous growth in access volume are intensifying the tension in computing resources. Behind the tension in computing power is a surge in energy demand. According to foreign media reports, ChatGPT’s daily electricity consumption may exceed 500,000 kilowatt-hours (500,000 degrees), equivalent to 17,000 times the electricity consumption of an American household. Another prediction also estimates that by 2030, the electricity consumed by AI will exceed household electricity consumption.
This is by no means an exaggeration; the improvement in chip performance, the layout of servers, the construction of data centers, network transmission, and other supporting infrastructure, even the temperature control systems for cooling servers all require substantial electricity.
With the enhancement of performance, the power consumption of Nvidia chips has also shown geometric growth. From V100 to B200, the power consumption of a single GPU has soared from 250W to 1000W, and the energy consumption of various supporting facilities corresponding to it has also surged significantly.
Moreover, before a large model is launched, it must undergo massive training. A Stanford study shows that the electricity consumption for training GPT-3 is 1287 MWh, while GPT-4 requires 51,773 MWh to 62,319 MWh (over 60 million degrees), which is more than 40 times the energy consumption of GPT-3.
If Google’s every search integrates artificial intelligence technology now, it would consume about 29 billion degrees of electricity annually. This is more than the annual electricity consumption of countries like Kenya, Guatemala, and Croatia.
In addition to basic electricity, the consumption of other various resources is also on the rise. Microsoft’s research shows that the water consumption of AI large models increased by 34% compared to the previous year, and total consumption of natural gas, electricity, heating, cooling, and steam has increased by about 32%.
Under the expectation that “computing power equals power,” AI is gradually becoming a core strategy for mainstream countries. It can be anticipated that in the future, multiple super large models similar to ChatGPT will emerge globally. Currently, AI is still in the initial stage of user experimentation, and large-scale applications have not yet been fully rolled out; if generative AI is used on a large scale, electricity consumption will inevitably rise sharply.
2. Abandoning Illusions
The construction of new energy far lags behind the growth of electricity demand.
Industry leaders unanimously believe that AI’s electricity demand is surging, but the pressure to reduce emissions is increasing, and the supply of new energy is not stable enough, while the outdated power transmission and distribution network exacerbates the supply-demand contradiction.
Similar to subway transportation, most of the American power grid facilities were built in the 1960s and 1970s and are very outdated. Among them, 70% of transmission lines and transformers have been in operation for over 25 years, and 60% of circuit breakers have been in operation for over 30 years.
Whenever there is extreme weather, power outages frequently occur. In July 2019, there was a massive blackout in Manhattan, New York; in August 2020, there was a massive blackout in California; and in 2021, Texas experienced a large-scale power outage.
Stimulated by the “Inflation Reduction Act,” the US is speeding ahead on the path of new energy, with coal power contributions significantly declining, while wind and solar energy are gradually becoming the main forces. In 2022, the share of coal power in the US dropped from 23% to 20%, and nuclear energy from 20% to 18%.
Currently, wind power generation in Texas has become the second largest power source in the state, but extreme cold weather has caused about half of the wind turbines to be “frozen.” The fundamental reason for California’s massive blackout is that the state’s implementation of the “100% renewable energy supply” plan is too aggressive, making the power system difficult to cope with sudden black swan events.
In addition, due to the lack of special steel required for transformers, the US basically relies on imports for transformers. Under the promotion of the Biden administration’s grid upgrade plan, the prices of transformers in the US have skyrocketed.
The vast new energy needs to be integrated into the grid; it is predicted that by 2030, the US transmission system will need to expand by 60%, and by 2050, transmission capacity will need to reach three times the current level.
Coincidentally, Europe’s power facilities are also outdated, and the EU’s aggressive clean energy strategy is incompatible with the traditional energy system.
For example, the UK government has promised to achieve 100% clean energy power by 2035. However, due to the lag in grid facilities, some solar and wind power generation needs to wait up to 10 to 15 years before it can connect to the UK’s power grid system.
To this end, the UK’s National Grid Company announced at the end of last year that it would increase investment by £2 billion, raising the investment for grid upgrades in 2025-2026 to £42 billion.
The integration demand for new energy has been severely delayed, while at the same time, the excessive demand for electricity from AI has already begun to emerge. An engineer from Microsoft’s data center predicts that by the end of 2024, when millions of H100s are deployed, their total power consumption will exceed the electricity consumption of all households in Phoenix, Arizona, and will rank fifth in the US.
Under multiple pressures, more and more people believe that unrealistic fantasies about new energy should be abandoned, and the value of traditional energy sources such as coal, oil, and natural gas should be re-examined.
As Saudi Aramco’s CEO Amin Nasser bluntly stated, the energy transition strategy has largely failed, and clean energy has yet to replace hydrocarbons on a large scale; the world should abandon the fantasy of letting oil and gas “retire” and instead make sufficient investments based on realistic demands.
The fact that energy infrastructure construction is far behind the explosion of AI has been increasingly recognized by more people. Thus, when Nasser’s remarks were made, they immediately received warm applause.
3. The End of AI
A fragile straw.
Currently, some believe that the internet and information technology have triggered a fourth industrial revolution. Even some opinions suggest that the outbreak of AIGC will bring about a fifth industrial revolution.
But the fact is that the second industrial revolution has not yet been fully realized; electricity has not been popularized globally. To this day, many regions, including some places in Africa, still cannot use electricity. Reports from the International Energy Agency and the World Bank indicate that due to factors such as the pandemic, the population without electricity globally has shown an upward trend again in 2020. If this trend cannot be reversed, by 2030, 660 million people will still be without electricity.
On September 4, 1882, Edison established the world’s first power plant in New York. In the following 142 years, humanity has actually long been in the process of the second industrial revolution.
Faced with the energy crisis, the constraints of aging power systems continue to hinder progress, and Europe and the United States are restarting traditional energy sources, especially coal, which was once scorned. Countries like the Netherlands, Germany, Austria, and Italy have had to reignite coal power plants that had already been shut down, which is essentially an acknowledgment of past radical energy transitions.
In the United States, the importance of renewable energy has also been highly emphasized, but faced with reality, there is no choice but to turn back to natural gas, coal, and nuclear power.
Most nuclear power plants in the United States were built in the 1970s and 1980s. After the Three Mile Island nuclear accident in Pennsylvania in 1979, nuclear power was gradually abandoned. Currently, only Unit 4 of the Vogtle nuclear power plant in Georgia is under planning, and actual construction has not yet started. Its sister unit, Unit 3, was connected to the grid last year, marking the first new nuclear power unit built in the U.S. in over 30 years. Due to a seven-year delay and a cost overrun of $14 billion, it may very well be the last glimmer of the U.S. nuclear energy industry.
Coal power has long been viewed as a source of pollution by the U.S. government, and reducing the proportion of coal power is an important task for the Biden administration. In 2016 alone, the U.S. shut down 531 coal-fired units with a total installed capacity of 56 GW. From 2022 to 2023, an additional 22.3 GW of coal-fired generation capacity was retired, and it is expected that by the end of 2030, more than 80 GW of power plants will cease using coal.
Currently, it seems that the U.S. can only rely on natural gas. Since the shale gas revolution, natural gas power generation has accounted for 40% of the U.S. electricity generation, making it the absolute main force.
When renewable energies like wind and solar cannot provide sufficient energy in the short term, natural gas remains the only reliable straw in the U.S. energy sector. However, in the face of the enormous demand for electricity from AI, this straw appears fragile.
4. Conclusion
Restarting coal, natural gas, and nuclear power is not a “restoration” of traditional energy, but rather an acknowledgment of the overly radical new energy transition implemented by many countries.
Traditional energy has made significant contributions to technological transformation but has been ruthlessly abandoned. We must also recognize that many traditional energy sources have long been clean energy across the entire industrial chain, while so-called “clean energies” like photovoltaics, wind power, and lithium batteries are not entirely clean and may even cause significant pollution.
The President of the European Commission, Ursula von der Leyen, once said: “Europe must ensure that it moves forward with this crisis rather than regress to dirty fossil fuels.” But this is just one aspect of energy industry development.
In fact, we should listen to the advice of Saudi Aramco’s CEO Nasser: “Only by enabling energy-poor countries to develop and utilize natural resources in a clean and efficient manner can we achieve fairness.”
This article is from the WeChat public account:Giant Tide WAVE (ID: WAVE-BIZ), Author: Xie Zefeng, Editor: Yang Xuran
The content reflects the author’s independent views and does not represent the position of Huxiu. Unauthorized reproduction is prohibited; for authorization matters, please contact [email protected]. If you have any objections or complaints regarding this article, please contact [email protected].
End
Want to gain knowledge? Follow Huxiu Video Account!