Tobold's Blog: The Terminator

Tuesday, September 16, 2025

The Terminator - 2025

A lot of people are afraid of artificial intelligence. They envision a future with elements from films like The Terminator, or A Space Odyssey, where AI becomes self-aware, super-intelligent, and intent on oppressing or wiping out humans. I don't think any of that will happen during my lifetime, armed robots simply aren't really a thing yet. In 2025 there is a possibility, albeit a small one, that you get killed by an AI-driven car. But we don't even have enough data, enough widespread adoption, and enough standardized AI driving technology that we could even say whether an AI driver kills more or less people than a human driver. The large language model generative AI from software like ChatGPT can kill people indirectly by causing mental harm, e.g. encouraging suicide, but that is still very far from a Terminator future.

In the medium term, a much more realistic fear of AI is that of a disruption of the jobs market through artificial intelligence. The nature of work will change somewhat if companies move towards a wider adoption of AI. New jobs will emerge, but some old jobs will disappear. If you never heard anybody describing his job as "Vibe Coding Cleanup Specialist", that is because that job didn't exist before AI. Today it does, but the company hiring him is probably hiring fewer regular coders.

I believe there is a much bigger short term risk of AI, but I will have to precede my thoughts with a "this is not financial advice" disclaimer. Stock markets are at an all time high. The S&P 500 is now over 40% tech stocks, and most of those tech stocks are heavily invested in artificial intelligence. There is at least a non-zero probability that there is an AI bubble, that investors overestimated the gains companies can make with AI over the coming years, and that the bubble could burst when these gains don't get realized as fast as they thought. How does that work?

What one needs to understand is the concept of market capitalization, market cap for short. For example Nvidia has 24.4 billion shares outstanding, and these shares today trade at $177.75. If you multiply these two numbers, you get the Nvidia market cap of $4.3 trillion. The problem with the concept is that this is imaginary money. It would be impossible to exchange 24.4 billion shares of Nvidia for $4.3 trillion in cash, because the act of selling Nvidia shares would make the share price drop. The same is true for every other share. As long as more people want to buy these shares than there are for sale, stock prices rise. As soon as people want to sell and cash in, stock prices drop. If you bought Nvidia shares on January 1st 2023 for under $15 and sell them now for $177, you made a huge profit. If you buy Nvidia shares for $177 now, and in a year the stock price dropped by half, you would lose money on selling.

Fundamentally, stock prices are correlated to what investors think the future of a company will look like. Recently OpenAI made a deal with Oracle to invest $300 billion in AI data centers built by Oracle. This raised the stock price of Oracle, but also of Nvidia, where Oracle would buy the chips for those data centers. If everything in this deal happens as announced, Oracle and Nvidia are expected to have higher revenue, and thus their stock prices rose on an expectation of that. But the future is uncertain, and this optimistic expectation of the future might just be plain wrong. OpenAI is actually losing money every year, and the $300 billion would have to come from investors. And then there is the practical difficulty of buying and powering that many data centers in just a few years. Once Nvidia has made the chips, sold them to Oracle, and Oracle has built the data centers, OpenAI still needs to figure out how to turn that $300 billion investment into a profitable revenue stream. With AI adoption by non-tech companies apparently slowing down, all that future profit isn't obvious.

I believe that stock markets are failing to price in the unintended consequences of the AI boom. For example, a lot of Americans have noticed their electricity prices going up, especially in regions where data centers have been built. Electricity is a regionally traded commodity, and if a lot of new electricity consumers go live, supply doesn't go up as fast as demand, and prices generally rise. That is a problem for AI data center owners, but also of all other industry using electricity, and of regular households.

Another "unintended" consequence of AI is more unemployment. Where do the stock markets think the profits from AI will ultimately come from? The most straightforward way for a general company to make money with AI is to replace people by it, paying less for an "AI worker" than for the person who did the same job before. But if everybody has "lower labor cost", it also means that a lot of people are unemployed or earn a lot less. And these people are the customers for the companies on the stock market. We already see companies like McDonalds struggling because many of their customers are lower income households, and can't afford fast food anymore. If you directly translate aggregate lower labor cost into aggregate lower consumer spending, the future for some companies on the stock market looks a lot less rosy. Those aren't necessarily the same companies, e.g. Oracle does very little business with low income households. But right now a glorious AI future, whether it will happen or not, has already been priced into the stock price of the companies that might gain from it, and not yet into the stock price of the companies that would be at the losing end of that deal. I don't see McDonalds replacing a lot of its staff by AI, as AI isn't yet very good at flipping burgers.

Investor optimism leads to overpriced stocks. "Irrational exuberance" in the words of Alan Greenspan. The price to earnings ratio of Nvidia is over 50, that of Oracle over 75. The historical average P/E ratio for the S&P 500 is between 10 and 20. Current S&P 500 PE Ratio is over 30. So the most short term risk from artificial intelligence is not some version of the Terminator, but what is politely called a "stock market correction", where investors realize that AI isn't going to pay out *that* fast, and stock prices drop a lot, because suddenly more people are trying to sell their overpriced AI shares than there are people wanting to buy them. And that also affects people who simply invested in ETFs, or who have money in a 401k pension plan. That is not to say that AI isn't a technology that will make money in the future, it just might take longer, or the profits will go to different companies than investors think today. Just like the dot com bubble burst, and the internet is doing fine, the AI bubble can burst while AI still has a bright future. The Terminator isn't likely to shoot you anytime soon, but he might still take down your retirement fund.

- posted by Tobold @ 10:37 AM Permanent Link

Comments:

There goes your pension money, meatbags!

# posted by

souldrinker : 16/9/25 11:18

Counter-argument: AI Continues To Make Rapid Progress (https://thezvi.substack.com/p/yes-ai-continues-to-make-rapid-progress), so the high valuations may be justified. This is from a subject matter expert who blogs on AI several times a week, not a random pundit jumping to conclusions about a hot topic. By which I don't mean you, Tobold, your reasoning is valid, as usual.

# posted by

Perkus : 16/9/25 18:36

I don't remember where I saw this article I'm about to try and mention so I can't link it but I saw something in the past few months talking about how the majority of companies that have been adopting AI tools are not seeing any increases or only marginal increases to productivity.

Obviously I'm sure the impact varies heavily from industry to industry and business to business but I personally think we are in a bubble and we will eventually see AI toolsets become more specialized and less a one-size fits all solution to anything and everything.

From my own personal experience an LLM specialized in a specific thing, like coding, generates far better results than just using one of the all purpose solutions or a plug-in. One of the vendors at my job sourced a plug-in with OpenAI specifically for coding and I do notice the performance difference when using their tool versus me just going to ChatGPT's public website and using the same prompts there.

# posted by

Bigeye : 16/9/25 21:05

I would agree more with tobold. The money needed to make AI profitable would either require every person on the planet to pay for it or perhaps charge 10x as much.

Ai has improved a lot in terms of cosmetic over the last couple of years but it still seems to fail on the same problems. Plainly put it looks better but it's not smarter. Your iq 60 chimp is now wearing a suit.

# posted by

Anonymous : 17/9/25 08:55

Armed robots very much are a thing. I mean no, not a humanoid one. But crawling, rolling and flying ones? They are wracking up more casualties than almost any other tech in the Ukraine war now.

The most commonly used long distance strike weapon in the Russia Ukraine war is now armed drones. The better ones now have an autonomous mode where they use AI to get to the target if they get cut off from communication with their home base, and can switch to navigating using on board gyroscopes if they get cut off from GPS. It's not as precise, but with high explosives you only need to get close.

There are also a large variety of various armed robots being used on the ground, though as far as I know none of those are any more autonomous than a robot arm in a factory or one of those robots used for disarming bombs.

In any case, there very much are armed robots being used in war and the tech is moving at startling speed. The basic doctrine of war is changing rapidly. Any military that can't effectively deal with swarms of low cost drones is going to be in trouble really soon. Million+ doller per shot missles are rapidly become obsolete for any but few specialized tasks.

For example, Ukraine has started to produce anti-drone drones, and that seems like the only air defense that is going to be low cost enough to remain viable to me. Using a 4 million patriot missle to shoot down a 4K drone is not econmically viable, and in a swarm of 100 drones it will only get a few at best. There are laser based technologies, but those don't work quickly enough to deal with a 100 drone swarm (unless you have like 20 of them in one spot). Robots using laser guided small arms fire is one good solution (think a low rent Vulcan defense turret), but I am not sure how far along that is.

# posted by

Yeebo : 17/9/25 17:38