Remember those news items that said these services were running at a loss?
Well, those calculations were based on pure compute costs, which DON’T include capital expenses like chips or hardware and it didn’t include labour costs which are, ironically, quite high for generative models
All of which means the economics for “AI” is much much worse than most people think.
Oh, and remember that the software industry is structured entirely on incredibly high margins (software being a non-rival, non-exclusive good and all). Even if the industry did manage to reach break even on “AI”, it would still collapse because it’s too dysfunctional to function on regular margins
There’s a rush in corporations to embed AI in everything. I wonder what happens when the AI companies start trying to raise prices. They are going to wait until it’s incredibly costly to remove these things. It’s going to be an interesting day when companies have to pay the full cost or yank all the integration. Just need to float it for a couple more years. It’s never going to get cheaper.
@jimgon@baldur already Microsoft is charging $30 per month and user for their Copilot for Microsoft 365 (the enterprise Office suite), despite not supporting too many languages yet.
$30 a month per user is cheap. And remember executives have been sold on unrealistic productivity improvement. $30 a month for the equivalent of 1/2 FTE?
But the current license cost seems like a loss leader. And if that is true, then the actual cost is going to be coming through after industry capture is complete. Or the whole scam collapses destroying entire industries in the process.
@jimgon@baldur But keep in mind this is one of a dozen different Copilot products. Copilot for Github? That’s extra. Copilot for security? Also costs extra. Etc.
@madeindex That 17x multiplier is Nvidia chips alone. So the hardware it’s installed in and the actual CPUs are extra. Research is extra. The army of freelancers used for RLHF training are extra. Electricity cost is extra.
And, fun fact, chips depreciate in value pretty rapidly. Esp since every chip vendor on the planet as more specialised ML chips in the pipeline that are more effective at the task. This investment will be worthless pretty quickly
@baldur I'm sure the big tech companies make sure to value the AI #hardware close to the original purchase price in their yearly financial reports, even after running 365 days @ 24/7 😂
CPU's / Graphic Chips probably survive the servers a maximum 3-10 years under these conditions right?
All it takes for the hype to be over is one generation of "*GPT" that does not improve, or doesn't improve enough.
Then the game is over and we'll see another AI-winter.
Yet, I am fascinated by people who derive such enormous benefits from AI having it write for them and code for them.
I have not been able to make an AI write or code for me with sufficient quality. In the end I always need to read through, modify and change, and in the end, it takes about the same time as if I
That is true. If we could predict when bubbles would pop, we would not sit here and chat on mastodon. We would be flying around the world chatting on mastodon from private jets. ;)
@baldur I don't want bad things to happen to good people, but I do want the current tech industry to burn to the ground, so that it can maybe be reborn into something better.
@baldur I have to admit, my Nvidia stock I bought a few years ago have appreciated nicely in value. The question is when I should sell. It's probably time soon.
@baldur@cstross A bunch of tech bros who have never had anyone tell them to keep things in budget - a whole generation that just spend other people's investments and deliver nothing.
That they also genuinely believe they can ask the AI to come up with ways to justify the capital expenditure is just the funniest part
Add comment