Do you guys think the AI/memory demand is temporary

[hhhhhhhhhhhhhhhhhh]

I have quite a lot of money allocated in those type of stocks

 

[lalalalala1234]

no it will def only go up.

 

[RJ_ascends]

I have quite a lot of money allocated in those type of stocks

holy npc
enjoy the crash

 

[proxxyy11]

no lmao it’ll keep increasing

 

[hhhhhhhhhhhhhhhhhh]

holy npc
enjoy the crash

Bro wtf youre 15 how would you know anything about stocks

 

[hhhhhhhhhhhhhhhhhh]

no lmao it’ll keep increasing

For some reason some people are comparing it to the 2000 dot com bubble but i feel like its different

 

[proxxyy11]

For some reason some people are comparing it to the 2000 dot com bubble but i feel like its different

completely different AI will obviously be the future and already essentially is. It wont stop as a trend dont listen to retards

 

[Rabbi]

It'll continue to go up.
That + military/defence related stocks are the best bets.

 

[eology]

It'll continue to go up.
That + military/defence related stocks are the best bets.

my ddr5 32gb ram costs same as gpu
last time i built my pc i let it fall by a meter tall, ocd trigger until now

 

[hhhhhhhhhhhhhhhhhh]

my ddr5 32gb ram costs same as gpu
last time i built my pc i let it fall by a meter tall, ocd trigger until now

Yeah its getting more expensive but the memory i meant was like the type of ram they use in data centers not really as much for pcs

 

[2ndAscension]

no lmao it’ll keep increasing

I have quite a lot of money allocated in those type of stocks

It'll continue to go up.

Chinese memory production is ramping up and consumer grade AI models are getting more efficient, 99% of AI tasks are served perfectly fine by something as cheap as GPT 5.6 Luna, and small open models like Qwen 3.8 27B trade blows with frontier models from half a year ago, but to make it clearer, look at these two benchmark graphs;

While Fable 5/Sonnet 5 in theory is the best available model, it's way too expensive to run for most users, and thus they have to economize their usage and make tradeoffs, instead opting to use a cheaper model, which does almost as good of a job. Extrapolate this economization to the industry in general, and it is likely that the far cheaper "almost as good" models will carve out a much larger market share in the future

 

[proxxyy11]

Chinese memory production is ramping up and consumer grade AI models are getting more efficient, 99% of AI tasks are served perfectly fine by something as cheap as GPT 5.6 Luna, and small open models like Qwen 3.8 27B trade blows with frontier models from half a year ago, but to make it clearer, look at these two benchmark graphs;
View attachment 5565807
View attachment 5565808

While Fable 5/Sonnet 5 in theory is the best available model, it's way too expensive to run for most users, and thus they have to economize their usage and make tradeoffs, instead opting to use a cheaper model, which does almost as good of a job. Extrapolate this economization to the industry in general, and it is likely that the far cheaper "almost as good" models will carve out a much larger market share in the future

love u tysm bhai

 

[hhhhhhhhhhhhhhhhhh]

Chinese memory production is ramping up and consumer grade AI models are getting more efficient, 99% of AI tasks are served perfectly fine by something as cheap as GPT 5.6 Luna, and small open models like Qwen 3.8 27B trade blows with frontier models from half a year ago, but to make it clearer, look at these two benchmark graphs;
View attachment 5565807
View attachment 5565808

While Fable 5/Sonnet 5 in theory is the best available model, it's way too expensive to run for most users, and thus they have to economize their usage and make tradeoffs, instead opting to use a cheaper model, which does almost as good of a job. Extrapolate this economization to the industry in general, and it is likely that the far cheaper "almost as good" models will carve out a much larger market share in the future

Right, so is your thesis that there will be more demand for data centers in the future or less?

 

[2ndAscension]

Right, so is your thesis that there will be more demand for data centers in the future or less?

Demand will increase, but supply chains will adapt and expand, while per token efficiency will increase, at some point an equilibrium will be met, and memory prices will decrease again
Remember how covid masks were extremely expensive for a short amount of time until supply chains and demand adapted? Expanding memory supply is obviously far more advanced than making a mask, but it will happen because there's money to be made, even when undercutting the main suppliers

 

[gooner23]

i feel like its too hedged on anthropic because they price they shit out of their models if anthropic can keep that business/government integration + cash flow not any time soon. but like there are literally FREE models just recently like ox alpha that perform near mid tier models which is where most tokens are burnt.

there will always be a demand for training hardware but those ai companies are making the chips themselves now to lower costs but i would still keep nvidia but sell anything thats not nvidia