Most super-tall skyscrapers are economic disasters. There seems to be a maximum economically rational height to a skyscraper, and it's already been reached. You can build higher, but if you do you're wasting your money.
A human-level or beyond AI would probably be like the Burj Khalifa: an economic disaster. Why build it when screwing and popping out babies is far cheaper and already works? If you want to exceed human intelligence, it would be a lot cheaper to augment human brains with external digital assistants (like what you're using now) or implants than to re-engineer an entirely new embodiment.
Why build a word processor when pencil, paper, and a scribe is far cheaper and already works?
It's about scale.
I assume you are familiar with this as 1988 = date of Fulkerson prize for Karmarkar's work? I guess the point narrowly holds if you're thinking of pure LP rather than general CP. But general CP is really quite a big deal.
As api mentions in the comments here on HN there are areas of work where progress stopped. For example passenger jet speed was once thought to continue at a rapid rate such that LA to Europe flights would be a few hours. Skyscraper height was thought to continue, with advances in various technologies and engineering methods making it desirable. Both hit realities that signficantly slowed down their progress.
I tend to side with Allen on this. While we're bright people, I don't know if I see us able to keep increasing computing power while keeping actual power consumption reasonably low.