I was looking up the resource requirements to run an "AI" model. It's quite expensive. A large model takes datacenter type infrastructure to run and can cost $50/hr or more to operate. The total amount to build one might be $500,000 to $1M based on some estimates. I'm not really sure how many users one model can service simultaneously, though, which is obviously a key metric.
Right now, there's a huge rush of investment into "AI", just like there was a wave of investment in network infrastructure in the dot com days. One of the effects of that investment is "AI" tools are more or less free to end users. The "AI" search results generated by google, for example, are free. Would anybody use it if they had to pay a monthly fee? I kind of doubt it.
The current gold rush model is predicated on the concept that every AI model will more than pay for itself, but that seems implausible.
No comments:
Post a Comment