- My Forums
- Tiger Rant
- LSU Recruiting
- SEC Rant
- Saints Talk
- Pelicans Talk
- More Sports Board
- Fantasy Sports
- Golf Board
- Soccer Board
- O-T Lounge
- Tech Board
- Home/Garden Board
- Outdoor Board
- Health/Fitness Board
- Movie/TV Board
- Book Board
- Music Board
- Political Talk
- Money Talk
- Fark Board
- Gaming Board
- Travel Board
- Food/Drink Board
- Ticket Exchange
- TD Help Board
Customize My Forums- View All Forums
- Show Left Links
- Topic Sort Options
- Trending Topics
- Recent Topics
- Active Topics
Started By
Message
The increasing power of AI Compute
Posted on 5/2/26 at 3:31 pm
Posted on 5/2/26 at 3:31 pm
I asked Grok to use 2024 as a baseline and give me the annual increase in AI compute for the next 10 years. There are caveats - natural restraints to growth such as available power, raw materials etc - but it's possible that through AI these constraints can be solved. Even if theses constraints slow the growth - the concept of half of this compute being available is mind bending. I don't think we're ready for the world this will bring...
Worldwide AI compute (primarily the stock of AI accelerators like GPUs/TPUs in H100-equivalents or equivalent FLOPs capacity) has grown extremely rapidly, at roughly 3.3–3.4× per year recently (doubling every ~7 months), according to Epoch AI data through 2025.
This measures the total deployed/installed base of AI-optimized hardware capacity worldwide, which best represents "compute power devoted to artificial intelligence." It includes both training and inference. Frontier training compute for the largest models has grown even faster historically (~4–5× per year), but the overall stock is the more relevant aggregate metric.
Projections (2025–2034, with 2024 as baseline = 1×)
These are illustrative projections based on continuing the recent ~3.3–3.4× annual growth rate from Epoch AI. Actual growth will likely vary and could slow due to power constraints, chip production limits, energy infrastructure, costs, and efficiency gains (though demand from inference and new models is expected to keep pressure high).
Many analyses project continued strong (though possibly decelerating) growth through 2030, with AI data center power demand and capacity potentially rising 5–10×+ in that period under high scenarios. I use a rounded 3.3× annual multiplier here for simplicity and consistency with recent empirical trends:
2024: 1× (baseline)
2025: ~3.3×
2026: ~11×
2027: ~36×
2028: ~119×
2029: ~393×
2030: ~1,300×
2031: ~4,300×
2032: ~14,000×
2033: ~46,000×
2034: ~152,000×
Cumulative growth to 2034 would be on the order of 100,000–200,000× under sustained exponential trends, though real-world physics (power, chips, data centers) will introduce bottlenecks and likely cause the rate to moderate, especially post-2030.
Worldwide AI compute (primarily the stock of AI accelerators like GPUs/TPUs in H100-equivalents or equivalent FLOPs capacity) has grown extremely rapidly, at roughly 3.3–3.4× per year recently (doubling every ~7 months), according to Epoch AI data through 2025.
This measures the total deployed/installed base of AI-optimized hardware capacity worldwide, which best represents "compute power devoted to artificial intelligence." It includes both training and inference. Frontier training compute for the largest models has grown even faster historically (~4–5× per year), but the overall stock is the more relevant aggregate metric.
Projections (2025–2034, with 2024 as baseline = 1×)
These are illustrative projections based on continuing the recent ~3.3–3.4× annual growth rate from Epoch AI. Actual growth will likely vary and could slow due to power constraints, chip production limits, energy infrastructure, costs, and efficiency gains (though demand from inference and new models is expected to keep pressure high).
Many analyses project continued strong (though possibly decelerating) growth through 2030, with AI data center power demand and capacity potentially rising 5–10×+ in that period under high scenarios. I use a rounded 3.3× annual multiplier here for simplicity and consistency with recent empirical trends:
2024: 1× (baseline)
2025: ~3.3×
2026: ~11×
2027: ~36×
2028: ~119×
2029: ~393×
2030: ~1,300×
2031: ~4,300×
2032: ~14,000×
2033: ~46,000×
2034: ~152,000×
Cumulative growth to 2034 would be on the order of 100,000–200,000× under sustained exponential trends, though real-world physics (power, chips, data centers) will introduce bottlenecks and likely cause the rate to moderate, especially post-2030.
This post was edited on 5/2/26 at 3:32 pm
16
Posted on 5/2/26 at 3:34 pm to Marquesa
Wow
This is what people don’t realize. Today it can’t but it won’t take a long time before it can.
This is what people don’t realize. Today it can’t but it won’t take a long time before it can.
Posted on 5/2/26 at 3:39 pm to Marquesa
People who aren't investing in these themes are missing out on a ton of money, it's not too late
Posted on 5/2/26 at 3:41 pm to Marquesa
But didn’t AI come up with this based upon what humans project AI to be? AI is still nothing without humans imo. Especially general use AI systems like Grok. Now I’m sure there’s some wicked AI systems that the public doesn’t have access to but still only good if humans use it.
Posted on 5/2/26 at 3:42 pm to Marquesa
Posted on 5/2/26 at 3:45 pm to DeBoar
quote:
But didn’t AI come up with this based upon what humans project AI to be? AI is still nothing without humans imo. Especially general use AI systems like Grok. Now I’m sure there’s some wicked AI systems that the public doesn’t have access to but still only good if humans use it.
AI can use logic and the available data (from humans) to make these projections. The point isn't that this is incredibly accurate, the point is that the increase is world changing. Even if the increase is 1/10th - then you're looking at a 15,000 fold increase in ten years. The world will be forever different - and no one knows what that will looklike.
Posted on 5/2/26 at 3:56 pm to Marquesa
In other words..... a fricking lot!
Call me crazy, but I think once an AI platform has achieves Super Intelligence, it will be somehow involved with the predicted anti Christ.
Call me crazy, but I think once an AI platform has achieves Super Intelligence, it will be somehow involved with the predicted anti Christ.
This post was edited on 5/2/26 at 3:59 pm
Posted on 5/2/26 at 4:07 pm to The Egg
quote:
People who aren't investing in these themes are missing out on a ton of money, it's not too late
What do you mean by "investing in themes"?
Posted on 5/2/26 at 4:17 pm to Marquesa
I for one welcome our future overlords and hope they give me a sweet gig in the coming matrix.
Posted on 5/2/26 at 4:20 pm to FluffyBunnyFeet
Data centers for AI
Power generation for the data centers
Semiconductors
Memory storage
Photonics
Etc
Power generation for the data centers
Semiconductors
Memory storage
Photonics
Etc
This post was edited on 5/3/26 at 9:49 am
Posted on 5/2/26 at 4:23 pm to lake chuck fan
quote:
In other words..... a fricking lot!
Call me crazy, but I think once an AI platform has achieves Super Intelligence, it will be somehow involved with the predicted anti Christ.
Mark of the Beast.
Posted on 5/2/26 at 4:31 pm to The Egg
Thanks!
Posted on 5/2/26 at 4:49 pm to Marquesa
I can't wait for Elon's universal income for all of us. AI does the work and we chill.
Posted on 5/2/26 at 4:51 pm to Marquesa
quote:
hough real-world physics (power, chips, data centers) will introduce bottlenecks and likely cause the rate to moderate, especially post-2030.
You have to build it.
Unlikely we could increase mining of raw materials by 40x in 3 years.
Posted on 5/2/26 at 5:08 pm to Marquesa
I'm just glad that they're going with the thin client architecture.
Posted on 5/2/26 at 5:18 pm to UptownJoeBrown
quote:
This is what people don’t realize. Today it can’t but it won’t take a long time before it can.
The leaps made from the beginning of the decade to now are stunning.
Posted on 5/2/26 at 5:18 pm to Marquesa
This is a projection based on if resources and capital were infinite. Neither are.
Posted on 5/2/26 at 5:49 pm to AUstar
quote:
I can't wait for Elon's universal income for all of us. AI does the work and we chill
Why would the elite let us live a life of luxury if they didn’t need us anymore. If that AGI future happens and we can’t make essential resources unlimited yet (post scarcity utopia) then our future is a boot to our throat and squalor if not an elite. You can see that is actually what elites are doing. Increased police militarization, surveillance from governments they control; private island and harden compounds for them.
Posted on 5/2/26 at 5:59 pm to Diego Ricardo
The main resource that matters is energy. With advanced technology and plentiful energy, most other base materials can be mined or recycled or harvested from asteroids, or synthesized from atoms. If we can solve the problem of energy, the scarcity issues "should" be manageable. That's my guess, anyway.
Posted on 5/2/26 at 6:00 pm to wdhalgren
Right I think AGI with fusion and quantum computers is that technological innovation that saves capitalism from itself once again. It’s been good at that which is why it’s been fairly resilient.
Page 1 of 2
Popular
Back to top
Follow TigerDroppings for LSU Football News