pressforge.io

For AI cost, there is an interesting scenario ahead:

Anecdotally, today when I code, and run out of Claude tokens, and have to pay more, i think to myself “this stuff is not cheap. It’s certainly not free”. In fact, in some cases it’s still cheaper to do some things manually.

Then, I think: wait, it will get 10x better in 1-2 years and you won’t have to spend that many tokens explaining, and correcting and iterating on simple things. This is why they are building all these crazy datacenters, is to make things more efficient, right?

Then: this is capitalism, these AI services will SURELY try to make much more money from you – personally you – over time, not less. To justify their valuations if nothing else.

And then I think, let’s say these AI tokens are heavily subsidized in 2026, and will 10x in cost after subsidies (VC funding) expire. If I spend say $500/mo on Claude et al today, if after subsidies removal it goes up to $5000/mo, then it’s cheaper and more efficient for me to hire a human or even two to do the same tasks. So the AI option just stops being realistic.

And the final thought, is if by the time the subsidies will start winding down, we as a society decide that this AI stuff is actually worth the trouble, then the market will figure it out one way or another, whether through cheaper open-source models that are somehow better distributed and don’t require such an enormous investment to continue, or again all these intergalactic data centers currently being built will start to kick in, and make things cheaper and/or more efficient.

How will this play out exactly? 🤔