All evidence indicates that we are in the final months of an era when computational resources are no longer considered the main constraint. It can be said that this is the final stage of the world where the scale of computations has ceased to be a bottleneck.
The largest manufacturers—ASML, TSMC, and Samsung—are operating at the limits of their capacities, and orders for gas turbines from all suppliers are scheduled at least until 2030. Investments in data centers in 2026 will surpass all investments made over the past thirty years combined.
There is still an opportunity to enjoy the moment when deploying state-of-the-art models is relatively straightforward—many believe: “Why run a simple model when I can choose something more complex?”
And for those who immediately want to comment: yes, of course, future models will become more efficient in parameters and require less active memory and computational resources. But it’s not just about algorithms or their optimization—the main issue remains resource scarcity. When major players like the Pentagon or Citadel are ready to compete for your GPU, you have to admit—your capabilities are limited, even if you have a couple of billion dollars in market cap.
Created with n8n:
https://cutt.ly/n8n
Created with syllaby:
https://cutt.ly/syllaby