Post by Christopher Mims / Redsky

shout out to @edzitron.com who is the whole reason my story this week exists -- he turned me on to the trend! go read Ed: www.wheresyoured.at/how-to-argue...

headline: Cutting-Edge AI Was Supposed to Get Cheaper. It’s More Expensive Than Ever. dek: With models doing more ‘thinking,’ the small companies that buy AI from the giants to create apps and services are feeling the pinch

aug 30, 2025, 6:27 pm • 617 64

Replies

It’s just been lies upon lies upon lies. Remember the arguments by the fanbros of “oh, give us like 15 minutes and we’ll ‘solve’ hallucinations?” You should dredge up all those confident proclamations and write a story on them. I’d love to see some good old trolling.

aug 30, 2025, 6:40 pm • 16 1 • view

This is a good idea. Comparing all the predictions of the past couple years to the current reality would give some perspective on new predictions.

aug 31, 2025, 4:21 am • 2 0 • view

I know you don't mean it this way. I can't help myself but I hate how hallucination became the word for what these models output. It is such a bullshit misleading marketing word. It's a valid but undesired output. There is no hallucination. The model functioned correctly.

aug 30, 2025, 8:37 pm • 4 0 • view

Agreed. If you knew me before we all became platform nomads I was a big proponent of "confabulation" to explain undesired outputs. But the morons, fanbros, and marketers absolutely insisted so I've long ago given up. It's bad enough I cling to "ML" to describe what we now call "AI."

aug 30, 2025, 8:47 pm • 3 0 • view

I think the correct term is "bullshit". Confabulation or hallucination are thought to be true. Bullshit is made up with no regard for fact or otherwise. The scientifically correct term is bullshit.

aug 30, 2025, 9:44 pm • 6 0 • view

Lol, true.

aug 30, 2025, 10:14 pm • 1 0 • view

that said, I am pleased that I got so many people to talk to me about the phenomenon of more-expensive-than-ever AI I dunno that this is an existential threat to all AI startups -- or any AI startups but only because there are macro trends that could be bigger threats www.wsj.com/tech/ai/ai-c...

aug 30, 2025, 6:30 pm • 125 15 • view

Them models be thankin'.

aug 30, 2025, 9:48 pm • 0 0 • view

Ya got class, Mims.

aug 30, 2025, 6:31 pm • 2 0 • view

speaking of "when will the AI bubble will burst?" Ed's latest on this is excellent. how much VC investment is left to prop up AI companies? 6 quarters. I have no idea if it will be a fast or slow deflation. there is real revenue here. just not enough for all www.wheresyoured.at/ai-bubble-20...

aug 30, 2025, 6:34 pm • 59 4 • view

I'm skeptical about "not enough" really qualifying as "real" revenue.

aug 30, 2025, 6:41 pm • 6 0 • view

I was a market analyst at a niche consultancy during the networks bubble. The scale of spending and the opulence of offices and events were nuts Demand was soaring but network rates were falling 70% annually, a pace no business plan could survive. Cue bankruptcy court and massive writedowns.

aug 31, 2025, 2:34 am • 1 0 • view

Writedowns enabled further price declines and ever lower prices helped spur a host of bandwidth intensive applications. The market found balance about 4 years post bubble. But investors lost almost everything, and fiber networks are still a very competitive, low margin business.

aug 31, 2025, 2:34 am • 1 0 • view

this is a great article. it would very helpful if you could shed light on Amodei and Altman's recently implications that their companies make money on inference. if that's true, then surely "everyone does a lot more inference" is a bullish sign rather than a bearish one

aug 30, 2025, 10:54 pm • 0 0 • view

here's the Amodei quote cheekypint.substack.com/p/a-cheeky-p...

Dario Amodei: Get kind of burned. There's two different ways you could describe what's happening in the model business right now. So, let's say in 2023, you train a model that costs $100 million, and then you deploy it in 2024, and it makes $200 million of revenue. Meanwhile, because of the scaling laws, in 2024, you also train a model that costs $1 billion. And then in 2025, you get $2 billion of revenue from that $1 billion, and you've spent $10 billion to train the model. So, if you look in a conventional way at the profit and loss of the company, you've lost $100 million the first year, you've lost $800 million the second year, and you've lost $8 billion in the third year, so it looks like it's getting worse and worse. If you consider each model to be a company, the model that was trained in 2023 was profitable. You paid $100 million, and then it made $200 million of revenue. There's some cost to inference with the model, but let's just assume, in this cartoonish cartoon example, that even if you add those two up, you're kind of in a good state. So, if every model was a company, the model, in this example, is actually profitable. What's going on is that at the same time as you're reaping the benefits from one company, you're founding another company that's much more expensive and requires much more upfront R&D investment. And so the way that it's going to shake out is this will keep going up until the numbers go very large and the models can't get larger, and then it'll be a large, very profitable business, or, at some point, the models will stop getting better, right? The march to AGI will be halted for some reason, and then perhaps it'll be some overhang. So, there'll be a one-time,

aug 30, 2025, 10:55 pm • 0 0 • view