In their defense, how many $20 billion fabs do you want to build in response to the AI ... (revolution|bubble|other words)? It seems very, very difficult to predict how long DRAM demand will remain this elevated.
It's dangerous for them in both directions: Overbuilding capacity if the boom busts vs. leaving themselves vulnerable to a competitor who builds out if the boom is sustained. Glad I don't have to make that decision. :)
I'd take the opposite bet on this. They're diverting wafer capacity from lower-profit items to things like HBM, but all indications are that wafer starts are up a bit. Just not up enough.
"Sequentially, DRAM revenue increased 15% with bit shipments increasing over 20% and prices decreasing in the low single-digit percentage range, primarily due to a higher consumer-oriented revenue mix"
(from june of this year).
The problem is that the DRAM market is pretty tight - supply or demand shocks tend to produce big swings. And right now we're seeing both an expected supply shock (transition to new processes/products) as well as a very sudden demand shock.
So it’s also the perfect time to constrain the product flow to jack up the prices.
They’ve been acting like a cartel for a long time now and somehow they never match the demand even after 18 months straight price increases. They already have the fab, the procedures, and everything, so stop acting like they’re setting up a brand new fab just to increase throughput.
This seems like a weird subject on which to be so aggressive, or at least I'm interpreting your tone that way. DRAM manufacturers absolutely have engaged in illegal price fixing in the past (1998-2002 in particular). But they've also overbuilt and underbuilt in fairly regular cycles, resulting in large swings in dram price and profitability. And they've had natural disasters reduce production capacity (e.g., micron in 2021). But there's no evidence right now that this is anything except finding themselves in the nice (but nervous) position of making a product that just just had a major demand spike, combined with some clever contract work by openai.
Demand right now is so high that they'd make more net profit if they could make more dram. They could still be charging insane prices. They're literally shutting down consumer sales - that's completely lost profit.
> I don’t think they’re working at 100% capacity or don’t have any other FAB that they can utilize for other low profit stuff.
I have a family member who works in a field related to memory and storage fabrication. At the moment Micron, etc, are running these money printers full time and forgoing routine maintenance to keep the money flowing.
What I said stands, let’s check their books and manufacturing schedule to see if they’re artificially constraining the supply.
The fact that they’re busy doesn’t hide the fact that they’re known to collude before, and they might even ship parts to phony resellers to keep the price high.
What’s next? A commodity memory chip is going to cost more than a cpu or gpu die?
Most of the things people say about efficient markets assume low barriers to entry. When it takes years and tens of billions of dollars to add capacity, it makes more sense to sit back and enjoy the margins. Especially if you think there's a non-trivial possibility that the AI build out is a bubble.
We only need the memory manufacturers to not collude with each other, not even external pressure.
You want to tame their cartel like behaviors? Just get into their books and it would be clear as day if they’re artificially constraining the supply, and I’m not even talking about spending extra billions.
You cannot manufacture something that modern life depends on and not get government scrutiny.
If it’s an AI bubble, it would be stupid to open new manufacturing capacity right now. Spend years and billions spinning up a new fab, only to have the bottom of the market drop out as soon as it comes online.
Assuming that opening a new fab is the only way to match the demand is simply asinine.
You can ramp up production in limited capacity, make long term contracts, or pass the manufacturing rust to the buyer. When we needed a vertical stabilizer for a legacy aircraft we paid for an entire production like to be built just to manufacture two tails, so there are tons of ways to do this if you want to be competitive. But instead this is a cartel like market where manufacturers colluded before, so they’re more likely to collude than spend billions doing anything.
Just open their books and schedules with a competent auditors and see if they’re artificially manipulating things or not.
A commodity hardware that’s on the price decline for decades just quadruples in price and nobody makes any form of long term investment or even contracts to take advantage of the situation? It’s more likely to be a collusion than not.
So much for open markets, somebody must check their books and manufacturing schedules.