Mastodawn

RE: https://hachyderm.io/@jenniferplusplus/116127090589590254

which is why there is no hoarding of all the computers and RAM for data centers that don’t even exist. what we are witnessing is a mafia-type collusion to remove from commercial circulation all the computational power people need to continue about their daily digital lives without paying rents to IP-thieving techbro clouds.

THERE IS NO RAM SHORTAGE.

THERE IS A BIG TECH COLLUSION TO CREATE A PC TECH SHORTAGE

memory is just the beginning. techbros are out to kill the PC market

Show thread

js Feb 25

@blogdiva Except that is not how it works. What happens is that HBM as used in datacenters has a MUCH higher margin than consumer grade DDR RAM. But capacities on wavers are limited. So those memory companies optimize for most money per area of waver. And since the margin for HBM is so much higher, they make almost no regular consumer grade DDR RAM at all anymore. So now the demand for the little DDR RAM that is being produced is vastly higher than what is produced, so there ends up a bidding war over the little DDR RAM that is being made. AI companies are not buying up DDR RAM and not using it - they’re buying the HBM and don’t care about your DDR RAM at all.

I hate the situation right now as much as everybody else, but spreading misinformation about it is not gonna help. If there’s anyone to blame, it’s capitalism and the memory companies having no reason to make DDR RAM with a lower margin.

// Edit: This is also the reason for the increase in SSD prices, because the companies that make memory and the companies that make SSDs are an almost perfect overlap. They just all only produce what is currently by far the highest margin for them.

Show thread

skry Feb 26

@js @blogdiva You mean chip wafers I guess.

I hope that this market opportunity causes Apple to make more fabs.

Show thread

@skry @blogdiva Building a new fab is a very long process (and Apple doesn’t have any, they just use TSMC). TSMC has been pushed hard to build more fabs, but they said no. Why? Because they anticipate that if they build a fab now, by the time they open it, the AI bubble has burst and there is not enough demand to utilize it and then TSMC would be sitting on the costs. There were some articles about how TSMC is sabotaging the “AI future”, how only TSMC is to blame for not providing more capacity, etc. trying to pressure TSMC into building more fabs. But from their perspective, it makes no sense. They need to plan long term - think 10 years and more.