Recently I spoke about AI with someone in private equity who argued that due to the costs of cooling data centers, the next logical step was to build data centers in space.
To me, this science-fiction viewpoint illustrates the current insanity of the AI investment bubble, where trillions of dollars are being thrown around haphazardly without serious thought about real world economics.
The Cost of AI
First, let’s consider the current and ongoing costs of AI.
Data Centers.
AI requires chips and energy for calculations and cooling, on a massive global scale.
Training.
AI requires massive amount of computing power and data to develop.
AI will also require significant power and data to maintain its knowledge.
Use.
Each AI query currently uses a surprising amount of energy, far more than a web search.
Eventually, AI should learn how to answer questions more efficiently at lower cost.
Next, let’s consider two different national approaches to building AI data centers.
USA
Billionaires are duplicating efforts in an unbridled egotistic race.
Data centers are currently an unregulated wild west, where costs to communities, water supplies, pollution, land use and energy use are minimized in pursuit of building rapidly.
This will result in political consequences and eventual regulation, increasing costs.
The efficiency of AI is also secondary to increasing scale: bigger is better.
China
AI development is a state run enterprise.
Political opposition is forbidden.
AI data centers are being built to be energy efficient.
AI models are being designed to be energy efficient by calculating watts per query.
How AI Earns Money
Infrastructure
Want bragging rights over your billionaire rivals? Just spend another 100 billion.
Corporate subscriptions
Want to replace another department? Just pay a quarterly fee.
Individual subscription fees
Want to keep chatting with your new ‘friend’ online? Just pay a monthly fee.
Who is Playing the AI Game
Jeff Bezos
Amazon Web Services hosts AI and also backs Anthropic
Bill Gates
Microsoft provides Cloud services and backs OpenAI
Page & Brin
Alphabet (Google) specializes in training AI, uses Gemini AI and also backs Anthropic
Mark Zuckerberg
Meta (Facebook) uses AI to milk its users and also trains AI models
Apple makes money from AI apps and cloud services. Nvidia makes money from AI chips. I don’t understand how Elon Musk plans to make money from AI, but he’s spending heavily on it.
How AI Economics Will Change
In the long run, technological competitions are won economically. The most advanced, first to market and even the most popular contenders will all lose if they can’t become profitable quickly enough and maintain higher profit margins.
AI demand will continue to rise. Think it uses a lot of energy to scour the entire Web to plagiarize your homework? Maintaining a steamy date with your VR sex bot will cost far more. As will replacing your accounting, legal and marketing departments. So will brainwashing millions of voters before the next election.
AI will destroy many careers and business models, destroying business profits and making many currently expensive services extremely cheap or free.
AI costs will drop. Ironically, future AI models will invent much cheaper ways of developing new AI models and serving queries much more efficiently.
The 1952 IBM 701 Electronic Data Processing Machine weighed 10 tons, used 20 tons of air conditioning, and took up over 1000 square feet. My iPhone has roughly a million times more memory and is hundreds of millions of times faster.—ECF
The Physics of Cooling AI
Inevitably, the data centers being built today will rapidly become obsolete due to their size, cost and energy use.
I don’t know exactly how the technology will change, but I do know how the earth revolves around the sun. Since night is cooler than day and winter is cooler than summer, one global solution would be to have data centers located around the world in both northern and southern hemispheres, calculating data sequentially all day and all year long in the coolest locations, while collecting solar energy during the day. That implies that no single country will host future data centers.
The Chinese operate an underwater data center powered by wind turbines.
Building (and servicing) data centers in space is obviously not economical.
Carbon fueled data centers contribute to global warming, so they must be stopped to protect life on earth.
The Bottom Line
AI is frequently wrong. I generally use it only when I already know the answer, at least roughly. The inaccuracies of AI are found everywhere online and unfortunately are being used to train new AI models in an error-ridden feedback loop exacerbated by exponential data center growth. Accuracy will become a competitive advantage, I hope, and I suspect that future, smarter AI will also be far more energy efficient than the current ‘boil the ocean’ approaches to data processing. Size matters not, in the long run.
Ultimately, overspending on inefficient, soon-obsolete data centers to rush into a world of rapidly diminishing profits is a risky investment strategy. Ultimately, investors demand profits not losses.
Doing so at a time when the US government is deeply in debt, a dependent Fed is being pressured to lower short term rates, inflation and long term interest rate expectations are rising, the economy is faltering, the stock market is historically overvalued compared to the size of our economy and political winds are growing against A) data centers, B) AI job destruction and C) corporate politicians, all seems particularly risky.
I don’t know when the AI bubble will pop, but I do know that the investment economics currently do not make sense. I explained my concerns to the private equity investment manager, but she explained that her qualified (meaning $5+ million) investors were throwing money at anything called AI. Which seems another excellent reason to tax the rich.
We’ll get straight to the point: The financial hardships that Daily Kos is facing this year are tough.
We continue to be paywall-free. We continue to be supported by our readers, not billionaires or corporations. But we need to bring in more revenue. We are leaning on our community more than ever to help make ends meet.
Power matters.