After reading Morgan Stanley’s key takeaways from NVIDIA’s NDR, here are some thoughts: 1. NVIDIA expects memory shortages to persist for several years. This is particularly noteworthy coming from NVIDIA—arguably the company with the clearest visibility into global AI demand and the most sophisticated supply chain operations in the world. 2. How should we interpret NVIDIA’s comment that “compute, networking, or memory shortages can be mitigated by optimizing the other two”? Is NVIDIA exploring new architectural approaches that leverage networking to compensate for local memory capacity constraints? CXL and memory pooling could potentially be part of the solution. 3. Has NVIDIA’s view of Groq changed? Previously, NVIDIA appeared to dismiss Groq’s SRAM-based architecture as suitable only for niche markets. However, recent comments highlighting the significance of Groq’s technology suggest a possible evolution in their perspective. 4. A more speculative thought: I’m growing increasingly bearish on Apple. I believe Apple could begin facing significant pressure starting in Q4 2026. If even NVIDIA is struggling with memory bottlenecks, it’s unlikely Apple will remain unaffected as shortages deepen. Apple and hyperscalers will undoubtedly develop technologies to reduce memory requirements. However, for hyperscalers, the benefits from efficiency gains may well be absorbed by Jevons’ Paradox. Memory is currently one of the greatest constraints on expanding compute supply within the AI economy. Therefore, if hyperscalers reduce memory requirements per unit of compute, the rational response won’t be to buy less memory—but rather to deploy more compute and reinvest the saved resources into purchasing even more memory. This dynamic will continue to push up memory demand from the AI server market, forcing competition over the same scarce memory supply and pressuring consumer device companies like Apple, which cannot benefit from the same AI-driven economics. Under this scenario, I wouldn’t be surprised if iPhone prices rise by as much as $400.