Exec Warns That Continued Price Hikes for Memory Will Become the New Normal, Lasting Well Beyond This Decade

[Peter_Brosdahl]

The current rampocalypse may be more than a short-run event, as one exec is now projecting its effects to be the new norm rather than the exception. A Lenovo exec speaking at ISC 2026 has indicated that projections regarding memory pricing are showing that price increases for memory components will continue into 2030 and possibly […]

See full article...

 

[screwtch02]

The fleecing will continue until the poors have none left to give....

 

[Brian_B]

I wonder if the AI overlords have considered if they continue to consume all the computer parts, eventually, who will be left over to consume their product?

 

[GodisanAtheist]

Fffffffffffuuuuuuuuuucccccccckkkkkkkk

 

[Grimlakin]

Fffffffffffuuuuuuuuuucccccccckkkkkkkk

My sentiments exactly.

Personally I hope these forward statements by CEOs lead to a massive increase in production capacity and the market is flooded so prices are forced to drop.

 

[DAPUNISHER]

My sentiments exactly.

Personally I hope these forward statements by CEOs lead to a massive increase in production capacity and the market is flooded so prices are forced to drop.

Not going to happen. Here is the manifesto. Avoid oversupplying the market. Expand production only when profitable demand exists. Don't repeat previous boom-and-bust cycles. Use long-term contracts instead of speculative production.

Just in time manufacturing and post human malware demand falling off a cliff was instructive for them. We will be the only ones getting shafted this go around.

 

[igor_kavinski]

The GPU hegemony will end as soon as a large enough LLM-on-chip is released. That will be a lot faster and if it gets 90% of the work done, there will be very few niche cases left to load up GPUs with LLMs.

LLM ASICs and specialized inference chips (why they matter)

Specialized LLM inference chips (Groq LPU, Taalas HC1, Etched, Cerebras) can beat GPUs on tokens per watt and latency. Benchmarks, architecture, trade-offs, and where to try ultra-fast small-model serving today.

www.glukhov.org

 

[Grimlakin]

The GPU hegemony will end as soon as a large enough LLM-on-chip is released. That will be a lot faster and if it gets 90% of the work done, there will be very few niche cases left to load up GPUs with LLMs.

LLM ASICs and specialized inference chips (why they matter)

Specialized LLM inference chips (Groq LPU, Taalas HC1, Etched, Cerebras) can beat GPUs on tokens per watt and latency. Benchmarks, architecture, trade-offs, and where to try ultra-fast small-model serving today.

www.glukhov.org

Sounds like.tje ant company will want to rule the roost.

 

[Skillz]

Similar to how ASICs killed mining BITCOIN on GPUs?

ASICs will kill LLMs on GPUs in the not so distance future?

 

[Marees]

 

[Marees]

View attachment 4699

apologies. apparently already posted in other threads

View attachment 4688

View attachment 4690

 

[Marees]

apologies. apparently already posted in other threads

 

[Ranulfo]

View attachment 4699

Heh, I hate that scene/section of that movie so so much. That meme amuses me greatly.

 

[Peter_Brosdahl]

Agreed and usually fun seeing it get reused over and over again for the most mundane of things.

 

[MadMummy76]

You know what? There has never been a bubble that hasn't burst.
web2.0 - gone
housing - gone
cryptomining - gone
nft - gone
metaverse - gone

They all thought they were going to the moon. So why would the datacenter bubble be any different?

 

[igor_kavinski]

So why would the datacenter bubble be any different?

Especially since they have built data centers that don't even have power available. It's major investor fraud. Promising them a future that the fraudsters have no intention of delivering. They will do the bare minimum, fill their swiss bank accounts with a fortune and disappear. Similar to a ghost city.