Toggle sidebar Toggle sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Apple Reportedly Agrees to 100% Price Hike on Samsung Memory Chips

Funny how the consensus is Samsung got the better part of this deal. I think the negotiation probably went more like -

Samsung - "How abut a 100% markup?"

Apple - "Sure, but we want 70% of your entire memory production, uninterrupted, for the next 5 years."

Samsung - 😳... "uh... let me make a call..."

Samsung got a deal in the short term. But Apple is playing the long game, because in 2027 and beyond memory prices will probably more than double, and Apple will have payed 2026 prices for their memory stock. Meanwhile everybody else is scrambling to find and afford memory supply.
 
  • Like
Reactions: zarmanto, ScanTheNavian and artifex
Delivered said:
Apple

A famously bad take. Running a fab for just yourself is a mistake AMD realized that and sold. Intel is learning that and trying to change the business model. This is like telling apple to start their own aluminum mine. 🤣
Click to expand...
That's my point. Maybe I misread but it sounds like you make a counterpoint but that's what say but with more words.
 
Ursadorable said:
The stock prices of companies heavily involved with AI have been consistently dropping as customers turn their noses up. Microsoft, Meta, Google, etc are all seeing significant drops in stock performance. I would argue that the bubble IS infact starting to pop. It's just going to take time for it to trickle down to the pricing on consumer goods.
Click to expand...

None of those companies stocks have been "constantly dropping" - neither are they AI only companies. They're primarily American companies and having much more impact are random tariff changes and horrific international relations.

The company really dependant on AI, Nvidia has held steady more or less since October, after seeing a near 50% increase in stock price last year.

The bubble is growing - and Apple has accidentally via Exo Labs become a big player in the local LLM game. We've sold more 512gb Mac Studio models in the last month than we did in the entire year they've been out previously.

I don't like that the consumer RAM and NVME's are now non existent because the largest companies in the world are only interested in producing HBM RAM - and that Open AI has bought nearly all of it with Google, Anthropic and Meta scooping up what is left - and because of this we'll probably see a massive price increase of the new 512gb Mac Studio when it comes up. People are snapping them up at £10,000 because it's a bargain for the RAM chips alone in todays climate.
 
bLackjackj said:
Over what? Apples worst negotiation ever? Gaaawd, whoever that was whom “negotiated” and accepted 100% should be GONE!
Click to expand...

You are assuming that that 60% price point was going to get Apple DRAM at the right time; as opposed to whenever Samsung happened to have enough left over to deliver. 100% probably includes time/priority guarantees ( most likely that Samsung has to hold some 'excess' production in reserve in order to adjust to Apple's requests. APple likely is going to still want 'just in time' delivery with their production schedules ( which are asynchronous with Samsung's) ).
If some other player neogiates later and pays the 100% and Apple is at 60% and there is question of who gets what out of the factory... you think Apple is going to 'win' in that context? Probably not. ( 'stuff happens' . Earthquake , storm, etc. batches go bad and there is a resynch then there is a reshuffle on priorities. ).


Furthermore. 40% less is NOT going to lead to new memory Fabs being built any faster. Paying less only means this shortage window lasts longer. There is such a thing as being penny-wise and pound-foolish. Part of the reason there is under capacity now is that Apple and other large players have been lowballing DRAM prices for a substantive amount of time.

As long as OpenAI continues to get 'drunken sailor' spending money from new/next cycle investors, Apple is in danger of being deprioritized. . Can Apple 'protest' about being deprioritzed. ( "we don't like that we are going someone where else". It is a completely empty threat. There is no untagged capacity to alternatively go to. Where is the negotiation here????? A game of 'chicken' where you sell less product when you come up short with no vendor? Samsung's 60% was ONLY a contingency in case Apple had already worked out a deal with someone else. Apple has the extremely crappy hand (playing Spades with zero Spades in their hand. ). )

Plus some of the M-series stuff is semi-custom RAM packaging. It isn't generic substitutable with random alternative vendor.

At this point. several of the large tech players with money are removing competition. Blowing up small phone vendors incrementally lets more of the larger players claw out wider moats of market share.
 
lotones said:
Funny how the consensus is Samsung got the better part of this deal. I think the negotiation probably went more like -

Samsung - "How abut a 100% markup?"

Apple - "Sure, but we want 70% of your entire memory production, uninterrupted, for the next 5 years."

Samsung - 😳... "uh... let me make a call..."

Samsung got a deal in the short term. But Apple is playing the long game, because in 2027 and beyond memory prices will probably more than double, and Apple will have payed 2026 prices for their memory stock. Meanwhile everybody else is scrambling to find and afford memory supply.
Click to expand...
One thing’s for sure, if I started a negotiation with a price expecting to go lower and my deal is accepted and negotiations over, the LAST thing I’d be thinking is “I’m such a good negotiator!” More like “My initial offer should have been at least twice as high.” I would have just left a meeting with some of the deepest pockets in the world and, instead of taking everything I could get, I left a fortune on the table.
 
  • Like
Reactions: zarmanto and lotones
Art Mark said:
Let's be clear here...NO ONE outside of Apple or Samsung have any idea what the deal actually is.
Click to expand...
Agreed, and the only hint is that one side is claiming a “win” through what’s they claim, was no negotiation at all. There’s two situations where people/companies are likely not to negotiate.
1. The deal is so horrible, it’s not even worth starting, and
2. The deal is SO incredibly good, even beyond your worst expectations, you want to get that number locked in and written in stone before they realize what they’re doing.
 
aloysiusfreeman said:
Vaccines have an astronomical amount more benefit to humanity than AI has given and will likely give.

Imagine AI on the battlefield, jfc.
Click to expand...
Imagine agent-driven vaccine/virus interaction modeling that can winnow down the molecular shape required for the vaccine to be effective against the virus. Imagine another agent modeling evolutionarily-beneficial viral mutations, and the agents go mano-a-mano for awhile until a vaccine turns up that will be effective against several viral strains over time, and vaccines that encourage evolution in a gain-of-function direction are winnowed out. Now the testing starts in the physical world, with many hundreds--or thousands--of ineffective or harmful candidates already eliminated.

BOOM--the time and cost of developing new vaccines has just dropped substantially.
 
swm said:
there's something fundamentally wrong with the world if a portable telecommunications/entertainment device needs 12GB of RAM (of any kind).
Click to expand...
Running language models locally (to preserve your privacy, improve latency, and reduce data center load) might well need it.
 
wikiverse said:
When is the bubble going to burst?
Click to expand...
You assume that there is a bubble. Yes, there's a lot of hype. Yes, there are a lot of companies getting investor cash by sprinkling "AI" in their pitches where it doesn't make sense, which will fail. Yes, there are a lot of companies that have good ideas that will not be able to execute, and they will also fail.

However, it is not for nothing that the juggernauts in the field are going full steam ahead, which they would not do if their forecasting told them not to (don't you think that multi-trillion-dollar companies have very skilled market researchers?). The people making hardware, in particular, who cannot pivot as quickly as the rest of the supply chain, are all-in.

Have you tried any of these tools in earnest? Once you see what they can do once you learn how to prompt them (and how to check their work), there is no going back. I asked one agent to research for me the risk of secondary cancer given a survivor who had a couple of particular kinds of therapy who--aside from the cancer--was in good health. In the span of an afternoon, with a couple of iterations, it had churned out a 30-page literature review, complete with annotations, that might have taken me weeks to do on my own. In my day job, I accomplished on Monday and Tuesday a couple of tasks that might have taken me a week apiece to do on my own, and further climbed the ramp of learning how to teach agents to do the tedious parts of the job for me in a repeatable way.

Just as with Henry Ford, assembly lines, and mass production a bit over a century ago and what value that brought to the human race, there is no going back.
 
citysnaps said:
If so... what an outstanding mistake that turned out to be!

Apple is now one of the most succesful tech companies in the world with 1+ Billion active and repeat customers who love Apple products and purchase them year after year after year.
Click to expand...

Please, Apple is no longer the largest tech company. Other tech firms have surpassed Apple under Tim.
 
  • Like
Reactions: JohnWick1954
Not sure on how much Apple will be able to maintain the pricing of the entry model. Maybe the base 18 could retain the $799 price tag but highly possible for 18 Pro and Pro Max to increase by $100.
 
  • Like
Reactions: mganu
aloysiusfreeman said:
Imagine AI on the battlefield, jfc.
Click to expand...
Why does anyone have to imagine it when it is happening and has happened for a while already in Eastern Europe? You might want to look into drone warfare in the Russo-Ukrainian war. AI has become a key part of it and it has turned conventional warfare upside down.
 
Let's see, Apple will likely raise prices "because of the increase in DRAM costs" and then in another 18 months when DRAM prices fall they will hold to that same increase. So, in the long run this is probably a win for Apple.
 
How many of these companies will find they’ve flown too close to the sun, raising prices so much that investors fund new competitors that steal all the long-term contracts once the bubble bursts? I’m hearing that Chinese RAM is competitive right now!
 
Keep in mind this 100% increase is only valid for one quarter. The next sit-down meeting will be even tougher.

People thinking prices will drop in 18 months are nuts. Hyperscalers have signed contracts through 2030.

DRAM suppliers are starting to force traditional customers like Apple to sign agreements that finalize pricing only after the memory has shipped. This is why Apple accepted the 100% increase with no questions asked.

Apple is paying more for RAM today and getting less. The prices for 2027 are expected to be worse than 2026. All of this negatively affects Apple. Higher prices means fewer iPhone customers. The DRAM suppliers are the beneficiaries, not Apple.

All of this is the same for NAND. Even HDDs are sold out through 2028.
 
extrachrispy said:
Running language models locally (to preserve your privacy, improve latency, and reduce data center load) might well need it.
Click to expand...
yes, I know that, I respect that, but sadly it doesn't seem to happen so. apple's transition to use privately hosted gemini might be just temporary - but the real challenge is that even 12GB isn't enough to run a modern LLM without serious optimization. and whether the temporary things becomes permanent, that's hard to tell.

and at the end of the day, the OS is getting more and more bloated, conveniently occupying the extra ram that was dedicated to the AI models.

don't get me wrong, I am all for on-device solutions, as that is the only thing what can scale. datacenters require a lot of investment, even more OPEX and constant upgrades. whereas you paid for your device, you charge it, and you'll upgrade when you think, you'd need more. also the combined compute power of all modern iPhones cannot ever be matched by no datacenter.

but I am almost entirely sure that the current "let's scale our LLMs horizontally" approach is a dead end, and likely new breakthrough(s) are needed in NNs/models (or their potential successors) to jump off from the plateau-ing efficiency curve. and as of now, the 12GB RAM seems a bit too luxurious for nothing.
 
  • Like
Reactions: extrachrispy
swm said:
yes, I know that, I respect that, but sadly it doesn't seem to happen so. apple's transition to use privately hosted gemini might be just temporary - but the real challenge is that even 12GB isn't enough to run a modern LLM without serious optimization. and whether the temporary things becomes permanent, that's hard to tell.
Click to expand...
LLMs aren't the only game in town. There are also "small" and "medium" language models which, depending upon what you need (just a better Siri that actually delivers on what Siri promised all those years ago?), might be quite enough.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back