They see the AI bubble for what it is, just like the rest of us, and they don’t want to be holding warehouses full of shovels when the gold stops coming.
yeah this is kind of the pattern with ram price fluctuations.
Ram demand goes up.
Ram prices go up.
Ram makers say they’ll increase capacity.
Nothing happens, they may open new factories but close older lines, or they may start to open another fab but then for whatever reason it doesn’t work.
The decision to build the new Boise fab in question (to start production in 2027) was made in September 2022. ChatGPT was made available to the public in November 2022. The new fab is likely not a response to significantly increased demand at the time but an investment made in expectation of increased demand (GPT-4 could already be tested in 2020, maybe they foresaw the LLM hype). They make DRAM. Isn’t it likely that prices are going to drop once production starts at the new fab?
Even massive efforts to build new chip fabs aren’t much help, he said, because the increasing complexity of new memory types means it takes longer to build factories – and when they come online there still won’t be enough capacity to build both the high-bandwidth memory needed for AI and other types of NAND and DRAM.
Isn’t it likely that prices are going to drop once production starts at the new fab?
depends on the market; one of the things I’ve seen repeatedly is new fabs opening to produce newer processes - replacing older fabs with larger nodes that then are shut down.
sometimes it’s advantageous to keep the old lines going and eek every bit of market share out of them, sometimes it’s prohibitive to keep older processes open.
but to respond to your query: in this market? in these crazy times? I’d be striking while the iron is hot and getting the maximum I could from every dram chip because the valuations of the hyperscalers and the surrounding ecosystem - open AI, anthropic, meta, google, nvidia etc., will continue to gobble it up until the bubble blows up in their faces.
coupled with the fact that the cartels refuse to expand production; this tells me they’re realistic about the moment - it’s not going to be a decade of future humongous peak RAM consumption, because otherwise they’d be blisteringly stupid (to lose out on those potential increased sales)… https://www.tomshardware.com/pc-components/dram/memory-makers-have-no-plans-to-increase-production-despite-crushing-ram-shortages-modest-2026-increase-predicted-as-dram-makers-hedge-their-ai-bets
They see the AI bubble for what it is, just like the rest of us, and they don’t want to be holding warehouses full of shovels when the gold stops coming.
entirely agree. if there were overproduction it would cause the prices to crash, can’t have that.
Micron has been building a new fab in Boise which is set to output RAM starting 2027, no?
https://www.kivitv.com/southeast-boise/everything-you-need-to-know-about-microns-massive-boise-manufacturing-expansion
Apparently another fab is planned already.
https://investors.micron.com/news-releases/news-release-details/micron-announces-groundbreaking-historic-new-york-megafab
Infineon is also looking to expand production, if I understand correctly, though RAM may not be their main thing.
https://www.stern.de/gesellschaft/regional/sachsen/chipindustrie--halbleiterproduktion--infineon-ist--ready--fuer-wachstum-37531970.html
yeah this is kind of the pattern with ram price fluctuations.
Ram demand goes up.
Ram prices go up.
Ram makers say they’ll increase capacity.
Nothing happens, they may open new factories but close older lines, or they may start to open another fab but then for whatever reason it doesn’t work.
Ram prices go up.
The decision to build the new Boise fab in question (to start production in 2027) was made in September 2022. ChatGPT was made available to the public in November 2022. The new fab is likely not a response to significantly increased demand at the time but an investment made in expectation of increased demand (GPT-4 could already be tested in 2020, maybe they foresaw the LLM hype). They make DRAM. Isn’t it likely that prices are going to drop once production starts at the new fab?
https://investors.micron.com/news-releases/news-release-details/micron-invest-15-billion-new-idaho-fab-bringing-leading-edge
Micron Locks In Historically High Memory Prices For Five Years
https://m.slashdot.org/story/455824
ho lee fuggin sheeeeeeeeeeeiiiiiiiiiiit
Even massive efforts to build new chip fabs aren’t much help, he said, because the increasing complexity of new memory types means it takes longer to build factories – and when they come online there still won’t be enough capacity to build both the high-bandwidth memory needed for AI and other types of NAND and DRAM.
Well, shit. So much for that. Is that even legal?
Another source with same info…
https://www.heise.de/en/news/Micron-sets-high-memory-prices-for-5-years-11344834.html
AI should be illegal if this is the consequence. fuuuuUUUUuu
depends on the market; one of the things I’ve seen repeatedly is new fabs opening to produce newer processes - replacing older fabs with larger nodes that then are shut down.
sometimes it’s advantageous to keep the old lines going and eek every bit of market share out of them, sometimes it’s prohibitive to keep older processes open.
but to respond to your query: in this market? in these crazy times? I’d be striking while the iron is hot and getting the maximum I could from every dram chip because the valuations of the hyperscalers and the surrounding ecosystem - open AI, anthropic, meta, google, nvidia etc., will continue to gobble it up until the bubble blows up in their faces.