The analyst is increasingly bearish on Apple, expecting significant pressure from late 2026 due to memory shortages impacting consumer-device companies like Apple more than hyperscalers.
Memory shortages are expected to persist, constraining compute supply. Hyperscalers can offset memory constraints by deploying more compute, but consumer-device companies like Apple face pressure due to scarce memory supply and may have to raise prices significantly.
A few thoughts after reading Morgan Stanley’s takeaways from NVIDIA’s NDR: 1. NVIDIA expects the memory shortage to persist for several years. This is particularly notable coming from NVIDIA—a company that arguably has the best visibility into global AI demand and one of the most sophisticated supply-chain operations in the world. 2. How should we interpret NVIDIA’s comment that shortages in compute, networking, or memory can be mitigated by optimizing the other two? Could NVIDIA be considering a new architectural approach in which networking is used to compensate for constrained local memory capacity? CXL and memory pooling could potentially be part of the answer. 3. Has NVIDIA’s view of Groq changed? NVIDIA previously appeared to dismiss Groq’s SRAM-based architecture as suitable only for a niche market. However, its latest comments emphasizing the importance of Groq’s technology suggest that its thinking may be evolving. 4. A more speculative thought: I am increasingly bearish on Apple. I believe Apple could begin facing significant pressure from 4Q26 onward. Even NVIDIA is struggling with memory bottlenecks. It is difficult to see how Apple can remain unaffected if the shortage becomes more severe. Apple and the hyperscalers will undoubtedly develop technologies that reduce memory requirements. However, for hyperscalers, any efficiency gains are likely to be absorbed by Jevons’ paradox. Memory is currently one of the biggest constraints on expanding the supply of compute, the most valuable resource in the AI economy. Therefore, when hyperscalers reduce the amount of memory required per unit of compute, the rational response is not to purchase less memory. It is to deploy more compute and reinvest the saved resources into buying even more memory. This dynamic could continue driving memory demand from the AI server market higher, while squeezing consumer-device companies such as Apple that must compete for the same scarce memory supply without enjoying the same AI-driven economics. Under such a scenario, I would not be surprised to see the price of an iPhone increase by as much as $400.
| Horizon | Raw move | vs SPY |
|---|---|---|
| 1 day | +0.6% | +1.4% |
| 1 week | +5.8% | +7.4% |
| 1 month | — | — |
| 3 months | — | — |
| 6 months | — | — |