It's much more bullish than just more CPUs for each GPU. I literally think the market can be worth trillions. Plus, the one company positioned the best BY FAR.
I am going to start putting the giveaway winner in the comments because it's getting insanely close and don't want to sow confusion if you were also very early. The winner this time was @doctor liquidity, congrats!
A devil's advocate to this is that intelligence (GPUs) and data (memory) can be scaled infinitely and limitless, whereas digital actions (CPUs) are somewhat capped. There is no need to perform more actions than what is required to accomplish the tasks on hand.
Which is why human civilization, innovation and progress are based on the scaling of intelligence and not actions.
Why did you sell $TSEM in April review? Also, Irrational said that highest performance compute is needed for AI, so I think $AMD is a play because of CPU Shortage + GPU hardware catchup, as CUDA gets chipped by AI software optimization over time
Pretty sure you're wrong about software efficiency gains, I expect 100-10000x is possible for most common tasks (somewhat less for relatively optimized stuff like databases, maybe only 10-100x). Most current software is horrendously inefficient & performance optimization is a easy task for LLMs
Getting the second half of OOMs here requires large redesigns though
With jevons paradox efficiency gains create more total demand. For example, 10,000x gain for AI memory dramatically increases text to video transition, AI agent proliferation, edge/inference AI proliferation. Memory has the biggest price hikes and efficiency gains help, not hurt them. Demand gap over supply widens, and is widening. Intel, AMD, NVIDIA and every hyper scaler (and sk hynix and samsung) have all said it just in the past month alone.
Micron is INSANELY undervalued, with a trailing PE ratio of 33x, forward PE ratio of 9x and PEG of 0.06. Let me say that again. PEG of 0.06. Ai memory is not commodity anymore. Three companies in the world make 95% of the HBM. No other company can possibly do it within 5 years. Micron benefits greatly from more CPUs, and ai agents and ai video, and ai reasoning, and edge ai and inference. And prices will continue to skyrocket. Micron signed a 5 year contract with a major customer. AI memory is not cyclical or commodity. It was always in the past but is not anymore. I've spent dozens of hours researching AI memory and Micron specifically. I like your articles a lot but I think you could use some more research on AI memory. I've created an indepth paper from all my research if you'd like to read. Again, i do like your articles a lot.
I am going to start putting the giveaway winner in the comments because it's getting insanely close and don't want to sow confusion if you were also very early. The winner this time was @doctor liquidity, congrats!
What is the investment rationale for going long SoftBank Group (9984.T / SFTBY) instead of directly buying ARM Holdings (ARM)？
OpenAI
A devil's advocate to this is that intelligence (GPUs) and data (memory) can be scaled infinitely and limitless, whereas digital actions (CPUs) are somewhat capped. There is no need to perform more actions than what is required to accomplish the tasks on hand.
Which is why human civilization, innovation and progress are based on the scaling of intelligence and not actions.
I like it
Pushback there will always be a cheaper and more active tier of AI carrying out tasks, not everyone is using Opus 25.0 to query databases all day
The fingers and brains was the best part of the article- very apt analogy! 😄
Why did you sell $TSEM in April review? Also, Irrational said that highest performance compute is needed for AI, so I think $AMD is a play because of CPU Shortage + GPU hardware catchup, as CUDA gets chipped by AI software optimization over time
Did you invest directly in SoftBank using a brokerage like IBKR, or did you buy one of the -Y or -F funds that track SoftBank?
Pretty sure you're wrong about software efficiency gains, I expect 100-10000x is possible for most common tasks (somewhat less for relatively optimized stuff like databases, maybe only 10-100x). Most current software is horrendously inefficient & performance optimization is a easy task for LLMs
Getting the second half of OOMs here requires large redesigns though
it's not clear to me how much optimization will actually happen though
Hey there, what do you think about the potential adoption of RISC-V-based CPUs?
Hi Jason,
Here is the link I promised that has my three reports on Micron. I recommend high strike price debit spreads, to get the most value.
https://we.tl/t-XJgz7Nv24tOwXdgc
With jevons paradox efficiency gains create more total demand. For example, 10,000x gain for AI memory dramatically increases text to video transition, AI agent proliferation, edge/inference AI proliferation. Memory has the biggest price hikes and efficiency gains help, not hurt them. Demand gap over supply widens, and is widening. Intel, AMD, NVIDIA and every hyper scaler (and sk hynix and samsung) have all said it just in the past month alone.
Micron is INSANELY undervalued, with a trailing PE ratio of 33x, forward PE ratio of 9x and PEG of 0.06. Let me say that again. PEG of 0.06. Ai memory is not commodity anymore. Three companies in the world make 95% of the HBM. No other company can possibly do it within 5 years. Micron benefits greatly from more CPUs, and ai agents and ai video, and ai reasoning, and edge ai and inference. And prices will continue to skyrocket. Micron signed a 5 year contract with a major customer. AI memory is not cyclical or commodity. It was always in the past but is not anymore. I've spent dozens of hours researching AI memory and Micron specifically. I like your articles a lot but I think you could use some more research on AI memory. I've created an indepth paper from all my research if you'd like to read. Again, i do like your articles a lot.