The AI cost issue is already rearing its head. People are using workarounds, and AI companies aren’t even charging a fraction of what it costs to run the models. What happens when they run out of investor money and need to make a profit?
Maybe the models will get cheaper to run?
I think that’s unlikely in the short to medium term. Current signs point the other way, as providers compete to make the biggest models and wrap them in traditional systems to create a commercially useful product.
Maybe the additional value that companies can get from using them will outstrip the price rises?
Again, unlikely in the short to medium term. There are impressive figures quoted about increased output, but outputs aren’t outcomes and not all outcomes generate value. When value is considered, the gains seem much less impressive. There are gains, for sure, but they’re measured in percentages rather than orders of magnitude.
So this really boils down to the investor appetite. Unlike crypto, there are real companies involved in this with real employees and real burn rates. Pump and dump in the bitcoin world just affects the price and the positions of those involved. Money can be made without the core asset making any kind of profit. All the players are doing is winning and losing in a closed system. Overall it’s a zero sum game (or negative sum game if you take the energy use into account).
The same is not true for AI, although some have tried to make the comparison. Here, investment is driven by an expectation of future gains, and those gains are priced into the market already. What if those gains are inflated way beyond what is actually achievable?
I suspect they are, which is a shame, because then the game becomes one of snatching defeat from the jaws of victory. A technology that could have been optimised and used to create value in excess of it’s cost might have delivered an ROI to all involved.
Unfortunately, the wild promises of AGI and 100x (or even 10x) productivity have pushed the stakes to unachievable levels. The companies involved are working hard to keep the dream alive by aiming for the next big hyped plateau without ever reaping the rewards of the previous ones. No-one is sweating the asset.
The amount of potential value that isn’t being harvested is hard to measure but it can’t necessarily be recovered further down the line. Certainly not by the incumbents because they’re riding on those promises and if reality bites, the investment money will disappear in a very loud pop. They’ll find themselves burning opex they don’t have and can’t easily shed.
The only answer will be to increase costs significantly or fold, and this is where the real 10x and 100x growth will happen. Not in productivity, but in price of service. Right now, AI is like a new drug. It’s flooding the market at knock-down prices and people are getting hooked.
When supply becomes scarce and prices sky rocket, it’s not going to be pretty.