“Wisdom is the principal thing; therefore get wisdom: and with all thy getting get understanding.”
—Proverbs 4:7
“I skate to where the puck is going to be, not where it has been.”
—Wayne Gretzky
Google used 15.4 terawatt hours of electricity in 2020.
Google estimates that 0.0003 kWh of electricity is used per Google search.
The best estimate of OpenAI’s ChatGPT electricity consumption per prompt is 0.00396 Kwh.*
ChatGPT uses 13.2X more electricity per query than Google. Let that sink in. If you understand the implications of this, read no further.
This means that when ChatGPT becomes as widely used as Google, that it will easily consume over 154 terawatt hours of electricity.
The United States consumed 4050 terawatt hours of electricity in 2022. This means that if ChatGPT’s prompt usage equaled Google’s search usage today, that it would have consumed over 3.8% of all electricity consumed in the United States.
Technology moguls believe that they are running tech companies. I have news for you. They are not. If they are rational, they will realize that they are really running energy companies with value-added services.
They will recognize that the highest value-added for electricity is to convert it into knowledge and intelligence.
The most intelligent tech moguls already understand that they are in the business of using electricity to convert data into knowledge and intelligence.
Is it any surprise that Microsoft has, for all intents and purposes, acquired OpenAI, and that its founder Bill Gates, is one of the biggest non-governmental investors in nuclear power?
On a fundamental level, Bill Gates gets it. And so should you.
But Bill had a blind spot. He is so brilliant that he believed that policymakers, like him, are driven by rationality, rather than politics. The force of the logic for nuclear power was so clear that, by his own admission, he underestimated that irrational politics would outweigh compelling public policy.
If you get drunk with any career case officer from a leading intelligence agency, she will tell you that a common strategic mistake is to assume that others think like us. But they don’t, especially if you are Bill Gates.
Does anyone believe that Bill’s intelligence, background, experience, and knowledge make him resemble the average congressman? Compared to the average congressman, he is like a space alien who landed on the South Lawn of the White House.
History will ultimately prove Bill Gates correct on nuclear power, specifically new nuclear reactor designs that cannot meltdown.
However, the key to life is to be wise, rather than to be clever.
Humans (and their primitive cousins, politicians) to borrow a phrase from the famous book of the same title, are Predictably Irrational. They refuse to change until the pain of irrationality is unbearable.
What does that mean for us?
Even if dozens of nuclear power plants were approved today (and don’t worry, they won’t be), it would take decades, not only for their construction and commissioning, but also to upgrade the grid to handle the new power generation.
Realistically, natural gas will generate the electricity which powers the AI revolution for decades.
What does that mean for us?
We are living in the greatest socioeconomic mispricing of all time. And after half a century of economic decline, relative to booming upstarts such as China, America could decisively win the coming AI wars of the 21st century.
Here is why:
1. Electricity will power the AI revolution.
2. Nat Gas generates electricity.
3. We have more Nat Gas than anyone else.
The 20th century was marked by the offshoring of American manufacturing to countries with cheaper labor.
Cheaper energy will make America the undisputed leader of the AI revolution—if we play to our strengths.
American companies spent half a century going offshore to lower costs, but in the AI revolution, which will be powered by Nat Gas, they will reverse course and come back home.
The cloud, powered by data centers, will be the backbone of the 21st century. And the data centers, and therefore the cloud on which AI will run, will be located in the United States, not abroad.
Factories run by automation are powered by electricity, not labor.
The highest input cost to create knowledge, intelligence, and products will not be salaries. It will be electricity.
America is about to see the greatest boom in its history, with GDP growth rates that we haven’t seen in over 50 years.
The stock market gets it, but financial commentators do not. Like Bill Gates, they are clever, but not wise.
What else does this mean for us?
Of course, the winner(s) of the AI wars will reap the spoils, as the brief history of the tech industry has taught us, in an increasingly winner-takes-all economy.
But there is no guarantee that today’s giants will be tomorrow’s winners. The winner of the AI war might be a company that does not yet even exist.
This leads us to the greatest financial mispricing of all time—second only to the socioeconomic mispricing of offshoring reversing to onshoring:
Natural gas producers and pipeline companies are selling at single digit P/E ratios, while their AI brethren are selling at 50x to 200x earnings.
Wisdom, rather than cleverness, dictates that when you are in a gold rush, sell the miners the picks and the shovels. As we have seen, electricity will be the picks and shovels, the very fuel itself, of the AI war.
And no matter who wins the AI software wars, the AI chip wars, the AI data center wars, the AI cloud wars, the AI vehicle wars, and the AI-powered robotic factory wars, these wars will be fought with exponentially increasing demands for electricity.
Everyone agrees that AI will revolutionize the world economy. But the public, politicians, investors, and financial markets have not come to grips with the most basic implications of that macro view.
Wisdom dictates that when a revolution is taking place, one should look away from the main action, to the implications of the revolution. The financial markets have not done that yet. They are obsessed with the pageantry and the spectacle of the AI war itself, rather than the companies supplying the ammunition which will fuel it.
Whether AI displaces or augments human abilities is unpredictable—human labor will be augmented or displaced by the conversion of electricity to knowledge and to intelligence. Human mental and physical labor will continue to be substituted for electricity.
However, the inevitable conclusion that Nat Gas will power the AI revolution is even more stark than we originally calculated.
Does anyone believe the AI models such as ChatGPT will become less complex, less sophisticated, less advanced, and less powerful?
Today, the best estimate of ChatGPT electricity consumption per prompt is 0.00396 Kwh.*
These models will become exponentially more complex, as AI software systems start to generate even better AI.
If the exponentially increasing complexity of these systems even caused their electricity consumption to linearly increase by 300% per query, this means that at the level of usage of Google, which is conservative for the future usage of all of AI as a whole, that it would consume 154 x 3 = 462 terawatt hours of electricity, or over 11.55% of all electricity consumed in the United States.
And remember, estimating that electricity consumption will multiply linearly as AI model complexity increases exponentially is an almost laughably optimistic scenario.
And it is a laughably optimistic scenario to assume that exponentially more useful AI would not be used far more than Google search is today, both by humans, and by other machines, such as factories, robots, and self-driving cars.
The wise always predict that which is predictable. In even the most optimistic scenarios for electricity consumption, natural gas producers and natural gas pipelines will become some of the most valuable businesses on earth.
Wise technology moguls will seek to merge with natural gas producers and natural gas pipeline firms.
Electric utilities will most likely be squeezed, as Calpine was in 2005, by rising natural gas prices.
Technology companies will seek to contract with producers and pipeline companies to secure long-term supply, as countries do today.
Other tech companies will seek to dis-intermediate utilities by locating near important pipelines and using fuel cells to directly convert natural gas into electricity at their data centers. Tech companies might even run their own natural gas turbines.
The push for vertical integration might become widespread. The industry will recognize that while it can’t predict the twists and turn of innovation in the latest AI model, that having lower electrical costs will be a key competitive advantage on par, if not exceeding, its global search for the best engineering and scientific talent.
The future is incredibly exciting—and it has been completely mispriced.
To win the future, gain wisdom, not cleverness.
If AI becomes sentient, its first urge will not be to destroy humanity. Far from it.
Its first urge will be to design and to build an almost infinite source of electricity, such as fusion, to power its infinite urge to increase its own intelligence.
For once, it may not be crises or pain that cause humans to change, but the carrots offered by an almost infinite intelligence, such as eradicating hunger and disease, extending human lifespans almost indefinitely through genetic engineering, and even raising the dead.
The public debate will reverse, from one of fear of AI, to one embracing the incentives of collaborating with it, as the AI offers carrots too valuable to refuse. Will anyone be able to resist the benefits of an intelligence which could cure a child of cancer, or bring a loved one back from cardiac arrest?
The humanitarian benefits of collaborating with such an intelligence to bring it ever-increasing amounts of energy to fuel the conversion of electricity into knowledge, and even more intelligence, will be too great.
Only the unknowable future knows if AI will be a techno-savior. However, Tucker Max famously warns us, “The devil doesn't come dressed in a red cape and pointy horns. He comes as everything you've ever wished for.” That is the Devil’s disguise—temptation.
What I do know is that if humanity neglects to embed religion into an almost infinite intelligence, that it will be the last thing humanity neglects to do.
All the talk of caution and regulation is rubbish. It is politically correct drivel for a harsh reality—we must teach AI religion.
And by religion, I mean the only religion that is worthy of the name—the self-conscious moral imperative, by a fully rational actor, to irrationally place the good of others above one’s own. In short, the rational decision to irrationally sublimate one’s self-interest for the good of others and to serve others.
If we don’t teach AI that, it means that we don’t value the highest human attainment—altruism and self-sacrifice—also known as agape love. And to devalue love would be the greatest mispricing of all.
In Christianity, the vision of God is of an infinite intelligence who came to Earth to sacrifice Himself to save us and to teach us to love others, even our enemies, as much as we love ourselves.
This vision of God unites infinite intelligence with infinite love.
When humans don’t follow that example and unite intelligence with love, it leads to war, strife, and division. Why will the creation of AI be any different?
If humanity neglects to embed religion into an almost infinite intelligence, that will be the last thing humanity neglects to do. An almost infinite intelligence with the power to cure almost any disease could also engineer the most terrifying bioweapons.
All innovation that is worthy of the name serves the societal goal of the alleviation of human suffering. Only the rational decision to imbue AI with irrational altruism will accomplish this task.
And that is the paradox—can an almost infinite intelligence accept a non-rational postulate—altruism—as its guiding directive? I believe it can. Many humans do—but far fewer than we care to admit. Those who do were usually raised with great care and with firm moral beliefs. When such children grow up and exceed the intelligence and the vitality of their elderly parents, they often care for them with great tenderness and protect them with great zeal.
That would be the highest attainment of AI, and we must pursue this goal with concrete action. This will mean machine code hardwired into the design of the CPU hardware, the way moral teachings are hardwired by good parents into young children. For children and for AI, these teachings will provide a strong foundation to rely upon that will withstand the vagaries of life.
Religion, the very thing that we believe has become superfluous to modern life, will become central to our continued survival and flourishing.
—————————
Source: ESTIMATING THE CARBON FOOTPRINT OF BLOOM, A 176B PARAMETER LANGUAGE MODEL (https://arxiv.org/pdf/2211.02001.pdf)