Hello, I'm Pazz.
There are a lot of news lately that Apple went to CXMT and begged for memory but was rejected..
Everything below is just my imagination, so please read it as fiction ㅎㅎ
Before the advent of AI, Apple was the kingpin of the DRAM market. There was no other device manufacturer that could absorb as much volume as Apple, and they were the biggest customer for Samsung, SK Hynix, and Micron.
When you make DRAM on a wafer, hundreds are made on one wafer. They test various things like functionality, performance, and speed on these wafers and categorize each DRAM chip as good - average - bad - dead. This is called Binning. At that time, Apple always demanded the best chips and the memory manufacturers actually did so. In other words, the best ones from the wafer went to Apple, the average ones went to Chinese smartphone makers, and the ones that worked but were buggy went to low-end set-top boxes, etc.
Apple even used its own unique packaging specifications for mobile memory, different from the general ones. Because of this, once a package was made for Apple, it couldn't be supplied anywhere else. If chips were tested to be slow in the packaged state, they couldn't be sold because they were in Apple's special format. This was the same for NAND. For memory manufacturers, it was a heartbreaking situation.
But with the emergence of AI, things suddenly went awry. Until last fall, there was no shortage of memory supply, so Apple could be picky about its memory purchases. However, with the surge in AI agent deployment and AI CAPEX, the memory shortage is expected to become very serious by 2026, and Apple's memory sourcing is starting to face headwinds.
It is said that the amount of mobile DRAM used in nVidia's Rubin platform, which will be released in the second half of this year, excluding HBM, will match the total global mobile DRAM demand. In other words, it means that one nVidia Rubin platform consumes as much memory as the entire smartphone market.
However, it is difficult to increase the memory supply by more than 20-30% annually for the time being. Therefore, the memory shortage is extremely serious and will be even worse next year.
Therefore, hyperscalers and big techs are entering into long-term supply agreements (LTAs) with memory companies for 5 years. As mentioned in the previous article, it is expected that the memory shortage will continue until around 2030, so big techs are also thinking this way and are entering into 5-year binding contracts.
In this situation, Apple is said to be asking for a price cut while still not having signed an LTA.
The problem is that memory companies have already fixed their supply volume for 60-70% of the total production by next year through LTAs. In other words, there is no more DRAM available to be supplied to Apple while they hesitate.
Apple, in desperation, went to CXMT. However, CXMT also had a problem because they were already supplying a lot of memory to Chinese big techs and couldn't spare any for Apple. Moreover, there was no way for them to meet Apple's special packaging and performance requirements.
CXMT's DRAM production process uses technology that is 3-4 years behind Samsung, SK Hynix, and Micron. The problem is that it consumes more power and is slower than the latest technology. If Apple used CXMT chips, the battery life of the latest iPhones would be significantly shorter and performance would suffer.
Rumor has it that Apple needs to launch new products in September, but they can't get enough DRAM, so the launches are being delayed (some rumors say they will be pushed back to early next year) or even if they do launch, there won't be enough supply. There are also rumors that TSMC has produced over $1 billion worth of Apple APs, but they have nowhere to go because there is no DRAM to attach them to. (Apple uses a process where it attaches its own special packaging DRAM next to or on top of the AP produced by TSMC.)
It's not only difficult to source memory now, but even if they succeed in sourcing it at a high price, it will take time to produce Apple's special packaging and attach it to the AP at TSMC. Therefore, it seems unlikely that new products will be launched on schedule in September. This is expected to have a significant impact on Apple's earnings.
Raising the price of memory would also lead to an increase in the price of Apple products, which could reduce consumer demand. It's not an easy decision to make. Galaxy, on the other hand, is minimizing price increases because they source memory from the same companies (although the DX division that makes smartphones has reported its first loss). In this situation, if only Apple raises prices... it will likely lose a significant portion of its market share to Galaxy. It's a real dilemma.
That concludes my fictional story.
Thank you for reading.
