groundzeros2015 12 hours ago

No? It’s an interchangeable component which is manufactured at scale by many suppliers.

Even though the elasticity of supply for taxis is less, rain encourages taxis to get on the road, and work longer to serve the spike in demand. With ride sharing apps the pool of supply is even more elastic.

▲

hackingonempty 12 hours ago | parent [-]

Building a RAM factory is a major investment and takes a lot of time. There is a big risk that by the time you enter production the rain will have stopped in the form of reduced demand and/or algorithmic improvements that reduce the memory required to produce good results. All of the attention is on the well funded frontier labs who may be buying up RAM as much to starve out competitors as anything else while in the background there is an army of researchers all over the world who only have a handful of consumer GPU to work with.

▲

groundzeros2015 12 hours ago | parent [-]

I already mentioned 3 ways we get more RAM and none of them require building new factories. Although, I would not doubt that effort is also ongoing.

▲

hackingonempty 12 hours ago | parent [-]

The only one you mentioned was existing factories extending production hours. AFAIK they already operate 24/7! Apple can't switch suppliers because everyone is selling out. Semiconductor factories are specialized and can't be easily switched to other types. It takes time and money and it stops making money for the duration, leading to a similar risk analysis as building a new one.

▲

Danox 5 hours ago | parent | next [-]

Apple can’t switch now, but they can take it in the house over the next 3 to 4 years to avoid this fiasco again. They have the right new CEO for the job. He’s a designer and engineer of chips, and since Apple didn’t waste money on the AI model/Data Center building exercise.

They certainly have the money to get it done in house, long-term, why? after this fiasco, the Chinese are going to have a much bigger piece of the memory market worldwide. So strategically, it pays to bring it memory in house in partnership with TSMC in Arizona or Oregon.

▲

groundzeros2015 11 hours ago | parent | prev [-]

1. existing factories increasing production 2. factories but are not making ram switch 3. Large consumers of ram use alternatives, broadening the supply

And let’s suppose none of these make a mark and a new factory needs to be built or something. This means: 1. You wait for build out and prices go down. 2. Prices go down anyway because demand is not sustainable.

And to turn it around, when you buy an expensive GPU to play computer games you are claiming a valuable industrial resource. Should the government subsidize your home consumption use case? Computer technology is a scarce resource with many uses.

▲

throw2ih020 10 hours ago | parent [-]

> existing factories increasing production

All existing factories have maximized their production.

> factories but are not making ram switch

It takes 2-3 years to switch, by which time demand may have satisfied from other manufacturers building additional capacity. So ironically, investing too much into new capacity can be dangerous.

> Large consumers of ram use alternatives, broadening the supply

What alternative exists for NAND flash?

▲

groundzeros2015 7 hours ago | parent [-]

> All existing factories have maximized their production.

Citation needed.

This is almost certainly not true because capacity is not binary but an efficiency curve. As the cost of RAM increases it becomes economical to operate the factory at higher capacities.

> It takes 2-3 years to switch

Citation needed. Who sets the max speed limit for changing?

> What alternative exists for NAND flash?

There is a whole range of suppliers. The alternative is which flash and who manufactures it.

▲

compiler-guy 6 hours ago | parent [-]

It takes about 3-4 years to build a fab and ten-billion dollars.

https://download.intel.com/newsroom/2022/manufacturing/fab-f...

And even if someone were able to magically build one in half the time, that would certainly drive up the cost quite dramatically, and would still be two-ish years from production.

The history of the memory industry is jam-packed with booms and busts, and companies that over-provisioned capacity during the boom times, only to have the bust happen as the fab is coming on line, are the ones that fail.

=-=-=-=

"William de Gale, portfolio manager at BlueBox Asset Management, told CNBC’s Europe Early Edition on Wednesday that the industry tends to have “enormous ups and downs”.

“In the long run it’s a pretty dreadful industry,” he said.

“I suspect that’s still the case every time people make an argument that the memory cycle is gone, and it’s now a long-term value-creating industry – just before it all goes horribly wrong.”

https://www.cnbc.com/2026/05/25/memory-stocks-cyclical-boom-...

	▲	Danox 5 hours ago \| parent [-]
		$10 billion dollars, 3 to 4 years, that’s less time to build a new modem chip that works. that’s less time to build a new M series processor that works. Google, Microsoft, and Meta We’ll spend close to $500 billion in 2026, just on their AI dreams, I would say having a supply of memory chips is vital towards your business if you’re someone like Apple or AMD or Nvidia, these days these days, if you want to design the devices you need to design, so who’s going to take it in house?