The squeeze
I hope the bottom falls out of the AI chip market and all your profits turn RED, greedy....*expletives*!!!
Memory-maker Micron has found a way to keep prices for its products sky-high for another five years, by signing 16 “strategic customer agreements” (SCAs) that include a floor price the company says comes with “a very robust gross margin for Micron, well above our peak quarterly margins in any past cycle.” Micron CEO, president …
Problem there is that...
If the AI bubble bursts and the AI companies are unable or unwilling to keep going at current levels, Micron might be able to say to (e.g.) Microsoft, "tough shit, you signed a contract, fuck you, pay me".
And if the contract was sufficiently watertight, they'll have to.
But with pure AI companies, that unlike Microsoft, don't have piles of cash elsewhere, have always operated in the red on the expectation of future returns and are still dependent upon a continued flow of investor cash, what happens if the bubble bursts and the cash stream dries up (or vice versa) and the value of their existing assets collapses?
Doesn't matter how tight the contract is, if no-one's willing to continue funding them, the money to pay those commitments simply won't be there. Creditors will be able to pick over the bones of a bankrupt company, but I doubt that'd cover a fraction of what was agreed to, even if future commitments were allowed to be prioritised over existing debts.
Over a barrel: you pay now or we sell it to you for even more in the future. (Assuming we decide to build the factory.) It's the price of locking in the price. You've got to be pretty confident prices won't rise further to refuse that deal. And if you're running an AI firm, you're not going to believe the bubble will burst.
Although, actually, if you think your AI company will survive the bubble then it still looks a good deal: because you're swimming in cash now. After the bubble, you may be struggling and not have the readies, no matter how much the price has collapsed
Apple has for many years done long term supply agreements where they paid in advance, to guarantee supply at a guaranteed price. They started this back when Cook was CFO when they converted the iPod from 1.8" hard drives to flash, and bought around 75% of the world's then yearly output of NAND to have enough to support that.
The problem is those agreements were rumored to be in the two year timeframe. If you assume Apple would do running contracts with multiple DRAM OEMs staggered say every six months, the last contract cycle where Apple would have got "normal" prices would be for H2 2025. The H1 2026 and H2 2026 contracts would come in far higher, and as the H1 2025 and H2 2025 contracts run out over the next year their average cost for RAM will match the market.
I'm thinking the massive hit they took getting the H2 2026 contract for the next two (or now maybe five) years is why Apple decided they could no longer hold the line on pricing anymore and were finally forced to do like everyone else already has and raise prices to account for the massive increases in DRAM and NAND pricing.
He also revealed that the SCAs see customers pay up front, which helps Micron to fund its fab expansions.
Only "helps", not pays for completely.
So if the bottom does drop out they are left with (possibly partially) completed fab expansions for certain kinds of complex high end memory but not enough demand. This is the classic problem undersupply/oversupply problem persistently plaguing semiconductor memory (and other parts) supply - and it kills.
I mean, it's all of the above for everyone, somewhere warehouses of thousands of robots are writing 'I will not murder Italians as an expat' with one hand while doodling new (but JEDEC compatible) DDR5 device plans with their right, middle, or other output mediating actuators (on paper, or merely low-refresh output devices.)
> I have seen numbers like these before. They did not end well...
> The most instructive parallel to today’s AI boom may be the railroad construction boom of the 1870s and 1880s. Like AI, railroads were a transformative new technology that promised to reshape the country’s geography and remake its economy.
> Source: https://fortune.com/2026/06/02/ai-bubble-history-railroad-crash-1873-pulitzer-liaquat-ahamed/
Though, I am in Europe mostly, let's all hope... we'll all manage it eventually...
If Microns gross margins are >80% and all the other suppliers gross margins are >80% then where is competition? I understand supply and demand but at those margins there is no incentive to increase supply or take the risk that in 3-5 years the demand will still be there. There is also no real incentive for new players in the market because of the 10-20 years it will take them to get to the standards required.
Lots of people talking about personal computing/devices but what happens to business computers/devices? They are in nearly every part of our lives now and a lot of them are going to need memory and SSD's and keep needing them. What's that going to do to inflation?
It's cooked. Capitalism has failed us comrades. The great eating of itself has begun in earnest. When the people can't afford what you make or provide who are you going to sell it to?
Two factors - inelastic supply and high barriers to entry. In short, the factory capacity for making Ram chips has a fixed cap until you build more factories which are VERY expensive and take a number of years to complete. Before that can happen, demand has risen massively (AI etc) much faster than supply can increase. As demand has increased and supply hasn't, prices rise.
Add into this that building a new factory may take longer than the AI bubble will last. It's possible that a new factory won't complete before demand plummet, so that's a further lag to increasing supply.
In reality every time you shrink a die or make a process improvement you increase supply in your existing fab. This thing is cyclic, In the early 90's there were maybe 20+ serious DRAM makers. Back then the 16Mb DRAM was hugely expensive and folk made a killing. That kicks off investment, increases capacity ..... prices drop, someone goes bust .... shortages ..... around and around it goes.
There is no doubt we are in a mega shortage, but we've been here before. Maybe the ramp of the Chinese will kill the current oligopoly of Micron, Hynix and Samsung.
It's cooked. Capitalism has failed us comrades.
It's not capitalism that's failed, it's the regulators.
Regulation is, and always has been, required in certain markets, otherwise people will be greedy, just like humans are wont to do.
If you think capitalism has failed your annoyed at an abstract concept with little recourse.
If you think of it as a failure of government then you can at least pester your elected officials to change it.
Mind you, that would require a good chunk of us not to just elect the person who promises the most, but think.
Damn you democracy!!!
I'm not sure you understand the nature of building DRAM factories. MU would love to sell more DRAM at these margins, but the leadtime on a new high-end DRAM factory is at least a year, and frequently quite a bit more than that. They are building new factories right now (check news release)
"I understand supply and demand but at those margins there is no incentive to increase supply"
Mmm... kinda. There's no incentive for them *all* to increase supply, but if only one of them does it, they get to reap the benefit of selling more (granted it'll bring the price down a little, but not a lot).
The problem each company has is if they *all* do it, because that would lower the costs back to normal levels, and negate the benefit. They can't afford to do it, but they equally can't afford to not do it in case anyone else does.
But the beauty is, they can do something like this - pre-sell to 16 chumps at an inflated rate and protect themselves from the latter example.
I don't think you understand what capitalism does. At 80% margins, there's plenty of reason to want to increase supply: if I sell 1000 items at 80% margin, but if I built a new factory I can sell 2000 items at 70%, I really want to do that because the increased quantity I sell helps the profit a lot more than keeping the margin up that high. This is the situation in a normal business in which capitalism pushes supply up quickly, and in fact it is sort of happening here too. The problem is that it takes a lot of time and money to build a RAM factory, and if I don't think people are going to be happy to buy at these prices in four years when it's done, why start building it now? It's not capitalism that's keeping factories from going up all over the place. It's the non-theoretical industry and caution, something most investors are cheerfully ignoring.
I do understand what capitalism does and you do have a point.
The problem is how much is that margin going drop? If I sell 20 of something for £4 and double the price to £8 but only sell 10 I actually just increased my profits because now I'm only making 10. I can also employ less people and pay myself more. It's a win win. That doesn't apply fully to the DRAM market but is it worth the risk for them? Capitalism is based on infinite growth and risk but the other side of it is to squeeze as much as it can for as little as it can hence enshittification. It's really not a great system for all but those at the top especially when those at the top dictate regulation.
Capitalism has plenty of problems, but none of the ones you've mentioned have much to do with the RAM market. RAM isn't expensive because somebody wrote regulation to benefit themselves. It's not expensive because they've got a plan to make your life worse to extract more cash*1. It's not expensive because they expect infinite growth*2. It's expensive because a lot of people want it but there isn't a lot of it making more right now is impossible, and making more long-term is risky.
*1 The most common conspiracy theory I've heard regarding the RAM market, though not from you, being that the RAM manufacturers are somehow colluding with computers to deny people local running capacity. It is ridiculous. RAM makers don't care what you do or how much other companies extract. For them to do that, they would have to be bought, and the price to do that, to prevent them from selling meaningful amounts of RAM for many years to come, is so much that even big tech couldn't afford it.
*2 In fact, if they expected infinite growth and "infinite risk" as you say (I'm not sure that makes much sense), the market would be easier as they'd have been building factories at a breakneck pace since two years ago and probably could have gotten investment from AI market promoters to do it. The supply is restricted specifically because they're smart enough to know the growth wouldn't be infinite and that, if they did build as much as they could get investors to fund, the bill would eventually come due.
Nice rallying cry, except that the big three Memory makers (and some of the smaller ones too) are actually expanding production capacity. And slightly faster that they would have in a counterfactual non-AI world.
¿Why don't they expand capacity even faster? They know that AI is in a hyped/hyper-bubble phase now, and they do not want to be holding the Bag when said bubble bursts or deflates...
Also, ¿Who said there is no incentive for new players? A new-ish player called CXMT from China is making the investments now to produce HBM3 in the near future, beyond their current line-up of DDR5/4, LP-DDR5/4. Currently only the big three are able to pruduce HMB3... And a old player in flash called YMTC is entering the DRAM space with LP-DDR5*, so, new plaers are actually entering...
But, do not let facts get in the way of a nice narrative. The greedy bastards are against us, the great eating of itself has begun, yada yada yada.
* Do not forget that, thanks to nVIDIA's and Micron's SO-CAMM (now overseen by JEDEC as SO-CAMM2) LP-DDR5 will be used for servers, and AI, beyond phones/tablets/laptops etc.
Perhaps if memory prices remain sky high for the foreseeable future it will encourage writing efficient programs again. Back in the 8 bit days with home computers with fixed, small amounts of RAM, developers would try to optimize and squeeze every last byte (and CPU cycle) out of a computer. For many years now computer software has been ever more bloated and memory hungry - in many cases unnecessarily so.
I fully accept the idea of not reinventing the wheel but the zeal with which libraries got dragged in to code has always concerned me. Especially when the coding avoided was trivial and those libraries often brought a fuckton of library dependencies with them.
But it was pointless to argue against "libraries, all the way down" bloat when memory was so cheap it was effectively free.
Mine's the hand-woven 1980's 'Told You So' T-shirt.
Probably less about the libraries and more about how people do allocations for data. From what I can tell, the browser guys are the worst. I've done chip designs and decent sized ones that use less memory to edit then a browser uses to show a cat video, like 50M device chips, with logic so not just a step & repeat pattern. The P&R people squeeze every bit, the timer people squeeze every bit, the DRC/LVS same... The browser people seem like it is a race to see who can use the most memory.
>” The browser people seem like it is a race to see who can use the most memory.”
I thought that was the webpage designers who from the outset competed to produce “dazzling” web pages, that could only be read by a user on the same LAN as the web server, at a time when the majority were accessing the web via 28.8~56kbps dial-out…
Except all the talented devs are earmarked for removal and what you will be left with are vibe coding marketing guys.
As I recall AI developed code is not known for being compact and efficient.
I can remember the days when they were hand writing games in assembly to save space. Seems unlikely we'll go back there.
Nah,
They've already been let go. These are the people who asked the awkward questions about this new fangled, greatest thing since sliced bread tech. The decision makers won't want them around when it all goes TITSUP and the brown envelopes stop coming.
True, but the simple fact of the matter is that graphics intensive games and user interfaces of the modern age require huge amounts of memory compared to the green screen era you're referring to, when software did only what it needed to for the specific job it was designed for, instead of importing the Universe knows how many graphics imports, transformations, and video embedding capabilities that are considered de rigeur for modern applications.
Back in the 8 bit days with home computers with fixed, small amounts of RAM, developers would try to optimize and squeeze every last byte (and CPU cycle) out of a computer.
Back in the 12- and 16-bit days of minicomputers with fixed, small amounts of core ...
Back in the 60-, 56-, 36-, and 32-bit days of mainframe computers with fixed, small amounts of core ...
But, yes. These days it's libraries, frameworks, and factories, all the way down.
Obviously I have no internal knowledge of the thinking behind Micron's decision, but there's something a bit fishy here.
If Micron believed that the demand for RAM would continue to rise over the next five years, they would be mad to agree to a price cap - in effect limiting their profits. In the event that things turned south [and depending on local laws] they might have some shareholder lawsuits on their hands.
Which suggests that they might have a different outlook.
There are a couple of possible explanations, but surely the most likely is that Micron suspect that the AI bubble will burst within the next 5 years and that this will result in a glut of RAM chips on the market. So by signing these agreements, they are in fact locking in their customers to what may become artificially *high* prices for another five years.
And of course Micron are in a position to know more about the marketplace than most, because they are the same company that walked away from retail sales in order to exclusively serve the biggest customer clients... They will be monitoring RAM sales to the bigger AI startup companies, so if they see a softening in demand, now would be the perfect time for them to strike deals that "lock in" the obscene profits they have been making.
It probably isn't a coincidence that this week's market sell-offs have been driven by investor jitters over whether the AI revolution will return profits or whether the valuations of the key stocks are over-priced.
This post has been deleted by its author
I think something is starting to shake at the highest levels. Musk is starting to do that thing where he starts ranting about government. This usually happens when his empire is starting to quake from financial issues he doesn't reveal. It has happened several times where after a year or two after the crises, he admits he was in a bind.
This post has been deleted by its author
You know, when the pyramid scam and stock market bubble of Artificial Ignorance bursts, the market is going to be flooded with RAM and those same companies are going to deeply regret these contracts.
I predict mass bankruptcies across the US and around the world when it happens.
Artificial Ignorance is just statistical prediction based on mass data scraping. It is not "intelligent", has no concept of "meaning", a "world view" model, a long term memory, or the ability to even keep track of more than one chat session at a time and learns not a single thing from prior sessions.
I've been experimenting with token optimalization towards my own MCP server and I found some chinese documents in the temp store and in skills dir.
They appeared after I asked it to optimize the skills for token effienency and compactness. (So even AI seem to prefere chinese).
the right time for a chinese company to sell affordable/cheap memory to the masses. They have the raw materials , just need to ramp up production and these days tend to just as good as the OEM's.
ChangXin Memory Technologies (CXMT) is one , I'm sure there's others that exist.
The other one up and coming is India , who have started chip production, way to go there before we see large scale manufacturing.
Micron's probably already worried about this.
> Why would China want to help the "rotten west"? It's a perfect opportunity to squeeze the "enemy of the communist revolution".
I would have thought the last four decades might suggest the PRC had screwed "the rotten west" by selling to us all the tat whose manufacturing we offloaded to them.
To be fair a self inflicted wound as much inflicted by our rich and poweful on our poor and stupid, as anything else.
Anyway we will impose a 1000% tariff on PRC RAM — that'll teach'em. /s
I would guess that the PRC will achieve tech parity with the west (by hook and by crook) in a decade — sooner if the AI bubbleburst's financial blast radius includes most of the world as it well might.
A few minutes thought of what a world with PRC parity with the west would look like makes the current geopolitical shenanigans appear even more ridiculous (and tragic.)
Even if the USoA and europe impose 100% tariffs on chinese DRAM, they will just sell to the rest of the world, meaning most of Asia, Africa, LatAm, the middle east (and china itself, of course).
This, in turn will mean less consuption of non-sanctioned brands in those terriories, which meanms more stock to sell in the EU and USoA, leading to a lower prices all around.
Since DRAM is more or less a commodity, the entrance of chinese manufacturers will mean lower DRAM prices, tariffs or no tariffs
China's YMTC, fine purveyors of NAND Flash storage*, announced in mid 2025 that they would enter the DRAM market with LP-DDR5.
And at a fantastic timing, because, thanks to nVIDIA's and Micron's SO-CAMM (now standarized as JEDEC's SO-CAMM2) LP-DDR5 can be used in servers and AI accelerators, instead of lowly phones/tablets/laptops etc.
Some other smaller purveyors (like Nanya Technologies) is also expandig production capacity of DDR5...
https://www.reuters.com/world/asia-pacific/nanya-technology-shares-surge-10-after-25-billion-private-placement-2026-03-26/
* No sarcasm here, at some point Apple qualified them as a provider for flash storage. Biden era sanctions (sustained by Trump) put a kibosh on that plan.
This post has been deleted by its author
The prices are unlikely going to go down.
Seeing margins this large, more players could join in based on that. They'll have no incentive to lower the prices as long as big corporations will have gullible and corrupt politicians as their clients with bottomless public purse and inept Chancellor ready to be wheeled out at moments notice that country needs more money and new taxes are needed.
You can tell all the hype and bubble is about to burst for multiple reasons, not least this deal. Primarily because all the AI companies are doing IPO and floating on the markets. They're effectively cashing in at what is hopefully (from their perspective) the peak.
The bad news for us is that the financial risk is now firmly away from the likes of Musk and venture capitalists - instead it's loaded into the stock markets, in turn the wider economy and your pension funds. They walk away with a fortune, which has effectively been borrowed from the economy.
And that will start to pop as the AI companies realise the "free" funding is disappearing and they need to make a return on their investment, and charge users more aggressively.
At which point companies reconsider how much they're using AI, and that there's a cost to giving all users access to CoPilot to have it write their emails in iambic pentameter.
Just the meter to compose the tragedies of our times. ;)
Myself, I would plumb for terza rima for a little renaissance elegance and flare… and just for the hell of it. ;)
"verso la fine del cammin di nostra vita / mi ritrovai per nella terra oscura / ché la diritta via era smarrita."
Very oscura and very smarrita, I should think.
The risk was never just with the venture capitalists, its also with the banks that have lent them money, and just with the sub-prime crash of 2008, they have sliced and diced those data centre loans up, and bundled them with other products which have spread to the whole global financial market.
If AI goes down, we all go down, but it will still be worth it to rid ourselves of this slop culture.
CXMT, Corsair, and Apple: The consumer DRAM revolt begins
CXMT's DRAM revenue surged 156% in 2025 to $8.6 billion, turning its first profit. Q1 2026 revenue hit $7.3 billion—up 700% year-on-year. It's now the fourth-largest DRAM supplier with 8% market share, and output bit growth will lead the industry again in 2026.
The Big Three—Samsung, SK Hynix, Micron—have pivoted to enterprise. AI needs HBM, margins dwarf consumer DRAM. The consumer market is under-supplied.
CXMT has no HBM contracts, no enterprise obligations. Its entire capacity is aimed at the consumer market the Big Three are abandoning. In LPDDR4X, CXMT could take 50% by 2026—it's already the largest supplier. DDR5 share is set to jump from under 1% to 7% this year; LPDDR5 from 0.5% to 9%.
CXMT isn't stopping. It's filed a $4.2bn IPO to fund R&D, building a new Shanghai fab with 2-3x Hefei's capacity, doubling monthly wafer output from 300k to 600k. It's expanding HBM production lines, targeting HBM3 by end-2026. It just signed a $3bn+ server DRAM deal with Tencent, with more Chinese internet giants in talks.
Corsair is shipping CXMT-based DDR5 kits—a major Western PC brand, CE-certified for Europe. Indistinguishable from Samsung or SK Hynix equivalents.
Apple is lobbying for permission to buy CXMT's DRAM, complicated by the Pentagon's blacklist. But Apple just hiked Mac/iPad prices, wiping $263bn off its market cap. The Big Three aren't making consumer DRAM for Apple anymore. Who else is he supposed to buy from?
Sanctions designed to slow China's semiconductor industry have accelerated its dominance in exactly the segment they don't cover.
CXMT is about to own consumer DRAM. And Western brands are queuing up to buy.
Seems everyone here is a consumer of memory and hates the huge DRAM and NAND prices. I understand.. However, Micron and their Korean siblings are not the ogre. The supply squeeze has been staring anyone in the eye for a couple years and anyone listening to earnings call's were being warned. Still no one paying much attention. Then the data center spurt is on and prices go crazy. Micron says they can only supply 60% of demand and had to appease long standing clients, new one's and say no to the lesser tier buyers (mid mobile OEM's) and others. For YEARS everyone basked in the over supply and and cheap prices, especially the chief scoundrel Apple which made more money on memory upgrade then the makers such as Micron. Apple pissed on Micron for years. Why on earth would the 3 Amigo's of memory keep overbuilding, over supplying and spending 10's of Billions on every new FAB only to make diddly squat on their investments. So now, think 2028 for any meaningful new supply and it will just trickle in as the poor souls not getting memory now can finally get some. China, not in the mix. Their market is domestic only. The smart play was to BUY equity in Micron, Sammy and SK Hynix. I'm a Micron long shareholder and sorry folks this buildup of under supply been poking everyone in the eye and easy to see if you were looking. Screw Apple who just can no longer screw their customers with over pricing the cheap memory they were paying for wielding their big stick which is now in hands of memory makers. The Hyperscaler's willing to pay ANYTHING to get their HBM is the real price driver along with 3x more wafers used up to produce it. Still, I just bought a new phone and lap top loaded with old price memory prices but that will soon be gone. Grab a beer it will be a while to buy at reasonable prices again...