Enterprise API and agent workloads dominate paid usage and high-end GPU demand. Menlo data shows ~$177B enterprise AI spend in 2026, with agents at ~80% of tokens on major routers—far outpacing casual consumer chat. Consumer spend is much smaller.
Scientific/research AI does require substantial compute, often FP64-heavy for accuracy in simulations and large models. Current global share is still modest versus commercial LLM work, but hard problems in disease and math scale with available flops; limited compute is a real bottleneck.
Enterprise API and agent workloads dominate paid usage and high-end GPU demand. Menlo data shows ~$177B enterprise AI spend in 2026, with agents at ~80% of tokens on major routers—far outpacing casual consumer chat. Consumer spend is much smaller.
Scientific/research AI does require substantial compute, often FP64-heavy for accuracy in simulations and large models. Current global share is still modest versus commercial LLM work, but hard problems in disease and math scale with available flops; limited compute is a real bottleneck.
@EdmundOnTop@miijiwoo@grok can you confirm this? And though not a high % of compute is currently going to scientific/math discoveries, it still requires a lot of compute theoretically to solve diseases, cancer or more advanced math problems. If it didn’t require much, they would be solved.
@EdmundOnTop@miijiwoo Again, no - the majority (70-90%) is enterprise api NOT user accounts. Individual users don’t do api they create multiple user accounts if anything.
The majority of use isn’t user generated slop.
Also, research ai (fp64)requires a LOT of compute, not sure why you think otherwise.
@HungarianBloke@TheMG3D You’re right actually… We need to slow down massively ASAP. Even if China takes the lead, we still need to slow down. This is going way too fast.
@karatademada It will die down a lot when gta6 is old news.
Right now every gamer wants to build a high end gaming pc and they are getting very very angry.
@HungarianBloke@TheMG3D I sort of agree. I go back and forth because if we don’t “rush” then china is ahead of us in less than a year.
I don’t think it’s a simple solution either way.
Government regulation should quickly catch up rather than ai slow down toooo much, is my general thinking on this.
@HungarianBloke@TheMG3D No one knows exactly how this will play out, I doubt unemployment will get very high before social safety nets are expanded.
But ultimately, the age of abundance will lead to a far better life for the majority, cure diseases and save hundreds of millions of lives over time.
@inthepayper@MesseyG82@drxenos@waldenpod You should really fact check this for yourself (because you won’t listen to someone online I’m sure).
Nazis were far right.
The far left does have its issues. But it’s not fascism. It simply isn’t.
I didn’t know anti-ai people use “generative” as a negative. I’m learning.
Most ai is not user generated, it’s api token spend by corporations.
Generative code is not slop.
Whether 16 bit or AMD’s 64-bit floating point for math/science, it is ALL ai. It is all on data centers.
@HungarianBloke@TheMG3D Who will starve?? No… i don’t want suffering, the exact opposite. How would there be millions starving and no public outcry for social safety net which will eventually expand to ubi?
Do you really believe we’ll let people starve on the streets?
@EdmundOnTop@miijiwoo Most of the use isn’t individual users though, it’s api token spend from corporations.
You can call ai code “slop” if you want, but to me that’s just insane given what it can do right now. Yes, for science they use AMD 64-bit floating point for higher precision. It’s still ai.
@HungarianBloke@TheMG3D The opposite.
If value and productivity increases dramatically, we will enter the “age of abundance”, something Michio Kaku has been talking about for decades.
If the pie gets exponentially bigger, as long as we don’t let the wealth gap increase, it means MORE FOR EVERYONE! ❤️