AI is two things that get treated as one#

Artificial intelligence is both a technology and an industry. My basic stance is that AI as a technology is compelling, and AI as an industry is the dot-com bubble on steroids. Large Language Models, especially open source models that are locally run, are compelling technological achievements. Commercial companies that are raising capital to bring LLMs to market are, in my opinion, doomed.

As an 80’s kid, I loved talking computers like in “Wargames” and talking cars like “Knight Rider”. I have always wanted an R2 unit or J.A.R.V.I.S. to help me with all the stuff I need to do. Having Max Headroom pop up on my TV would be awesome. Especially if he and K.I.T.T. hated each other. I love the idea of voice and natural language as an interface to a computer. Is AI there yet? Not even close. Alexa isn’t any closer to R2D2 than a Meta Headset is to jacking into The Matrix.

The VR headset isn’t going to revolutionize work or entertainment, and neither will the Large Language Model. So with all of that in mind, here are my hot takes about AI:

Take 1: The data center issue doesn’t make sense to me#

I have worked as a data center tech. My experience with data centers is that what you spend on servers you will spend again on network equipment, and what you spend on servers and networking, you will spend again on power and cooling. Roughly $1 million in compute and storage requires close to that again in routing and switching, that $2 mil needs at least another $2 mil to power and cool it.

One of my last IT jobs before retirement was for a small company whose director of IT built the computer room himself, using commercial air conditioners (mini splits). The generator was sized based on the power needs for the servers and network equipment. When a big storm came we had to power off the entire computer room because the generator wasn’t going to handle both powering the equipment and keeping it cool.

All of this is to say that cooling is a massive expense. These news stories about data centers the size of Manhattan being built in the desert don’t mesh well with my understanding of how data centers work. Commercial data centers are built around cooling via Computer Room Air Conditioning, or CRAC units. The Iron Mountain facility in Pennsylvania is built into a limestone mine. The thermal mass of the rock and the underground lakes are part of the cooling system. European data centers get built in places like Ireland and Sweden because the ambient temperature is cold most of the time, and they are less freaked out about nuclear power. A desert in the American southwest only makes sense if you are getting the power for incredibly cheap, like hydroelectric, or solar, because you will need a lot of power for cooling.

The thing with water cooling systems is that they’re closed loop systems. You pump water through heat sinks and then cool it with a radiator. Why is there a threat to the local water supply? If anything, I would worry about the local supply of diesel or natural gas, because that’s what backup generators would run on.

I’m not saying anyone is lying, just that what is being said doesn’t line up with what I understand data center engineering to be.

Take 2: This is the dot-com bubble all over again#

I lived in Seattle from 2000-20002 during the dot-com bubble. Red Hat went public and was valued at a billion dollars. There were Pets.com ads during the Superbowl.

Fast-forward to 2026: Nvidia is worth a trillion dollars and Anthropic running ads on the Superbowl.

The dot-com crash had a significant effect on my life. I went from living on the West Coast bossing around dudes with master’s degrees to living in my mom’s basement and working at a help desk. It took me a few years to get back to where I was before I moved out west. This time around I didn’t just not go out west, I retired from IT. Fool me once, etc.

During the dot-com bubble, there was pressure to buy servers, routers, and broadband. WorldCom and Enron turned out to be frauds. Now it’s memory and graphics chips, and I am waiting to find out which companies are frauds.

Take 3: The value proposition of AI is eliminating wages#

Every CTO that is pressuring their employees to use AI is under pressure from their CEO to eliminate employees. It’s not going to work, but you can’t blame them for trying, a world without workers is every neoliberal’s wildest dream.

This is the same promise that shipped manufacturing to China and tech support to India. The lower cost product is inferior. The AI taking orders at the Taco Bell drive thru is somehow a worse experience than ordering from someone out of their mind on drugs. At least you can reason with someone on drugs.

What the value proposition should have been is outsourced emotional labor. AI is great at explaining Linux video driver errors to me. It doesn’t get upset explaining the basics of social justice. It’s like Googling a word to see how it’s spelled, but more involved. Is it a replacement for interacting with Linux users or feminists? No, but the Linux users and feminists don’t need to waste their time referring people to the FAQ. AI is an FAQ that reads you instead of the other way around.

Take 4: LLMs are now what Wikipedia was 15 years ago.#

When I was in grade school, I was told to not use an encyclopedia for research papers. But if I was writing about something I had zero knowledge of, an encyclopedia was a great first step to getting the general idea. It was like reading an outline, that I could flesh out with actual research.

Wikipedia was the encyclopedia with search and hyperlinks. An LLM is Wikipedia that you can ask follow-up questions. It’s not going to make the children illiterate. It might impact their critical thinking, but you can configure it to help with that too. Just put “don’t shatter my critical thinking” in your soul.md file :-)

I don’t use AI to do my writing for me. The whole point of writing for me is handling depression by doing something creative that I slowly get better at. There’s no point in outsourcing that work.

Take 5: I would gladly live in a Terminator-style hellscape if I could have an AI Dungeon Master.#

I know I am supposed to be resistant to big tech and protect personal freedoms. But getting a weekly D&D game that actually happened on a weekly basis? That would be pretty dope. Can the Human Resistance guarantee my game actually happens? Maybe they should keep that in mind for recruiting purposes.