Apple’s iPhone 18 chip plans may be hitting an unexpected speed bump, and the culprit isn’t performance or manufacturing yields—it’s memory. Ongoing DRAM shortages and rising prices are reportedly forcing Apple to scale back a key packaging upgrade that had been rumored for the standard A20 chip expected to power the base iPhone 18. The situation is a reminder that even the most supply-chain-savvy tech giant can’t fully escape the volatility of the global DRAM market.
Why the A20 packaging change mattered in the first place
Until recently, expectations were that Apple would move the A20 away from TSMC’s InFO (Integrated Fan-Out) packaging and adopt a newer approach called WMCM (Wafer-level Multi-Chip Module). That shift wasn’t just a technical makeover—it could have unlocked more flexible chip layouts and, crucially, better memory performance for demanding on-device AI features.
WMCM is designed to combine multiple separate dies—such as CPU, GPU, and the Neural Engine—into a single compact package. That opens the door to a wider range of configurations, including different mixes of CPU and GPU cores depending on the product tier. Another benefit is efficiency: with WMCM, the CPU, GPU, and Neural Engine can behave more independently, drawing power more precisely based on the task at hand, which can help keep power use in check.
But the biggest draw for iPhones was always memory. WMCM can place RAM directly onto the wafer alongside the processor rather than sitting adjacent to it. That on-wafer placement improves latency, and lower latency can translate into smoother performance in AI-heavy workloads, where rapid memory access matters. WMCM can also improve heat management and increase interconnect density by placing critical components on a redistribution layer and avoiding certain extra layers used in other designs.
So why back away now? DRAM pricing and supply pressure
According to a recent industry tip, Apple may be dropping WMCM for the standard A20 because the current DRAM shortage—plus higher DRAM prices—makes the broader strategy too expensive to roll out across the lineup. The goal behind WMCM for the base A20 was reportedly tied to enabling more DRAM with better latency for edge AI. If the memory is harder to source or significantly pricier, that plan becomes much more difficult to justify in the lower-margin base model.
What happens to the A20 Pro and iPhone 18 Pro models?
The same report suggests Apple will still use WMCM for the A20 Pro chip, which is expected to power the iPhone 18 Pro and iPhone 18 Pro Max. However, there’s a catch: even with the more advanced packaging, those higher-end models may not get an increase in RAM capacity.
Instead, Apple is said to consider today’s 12GB LPDDR5 memory sufficient, allowing the Pro models to gain the packaging-related advantages (like latency and efficiency improvements) without the added cost of increasing RAM at a time when DRAM remains expensive and constrained.
Bad news for base iPhone 18 RAM expectations
If WMCM is indeed being reserved for the Pro chips, it also makes a 12GB RAM upgrade for the base iPhone 18 far less likely. Recent speculation had pointed in that direction, but the economics don’t appear to support it. The cost of a 12GB LPDDR5 module is projected to be extremely high around the timeframe when the base iPhone 18 is expected to launch, making a meaningful RAM bump hard to justify for Apple’s standard model without squeezing margins or raising prices.
The bigger takeaway: AI ambitions meet real-world component costs
Apple has been pushing more on-device intelligence, and memory is one of the most important ingredients for AI features that run locally—especially as models and workloads grow more complex. But this report highlights the practical limits of that ambition when supply and pricing don’t cooperate. If DRAM remains tight and expensive, Apple may continue to differentiate iPhone 18 features not just by camera upgrades and display tech, but by behind-the-scenes silicon and memory decisions that determine how far on-device AI can go on each model.
