HBM growth attributes benefit GPU leader

Is there no cycle for HBM anymore? Traditional DRAM follows the end market: how many PCs are sold, how many phones are shipped, how many servers are bought. Growth is relatively linear; when the cycle comes, everyone falls together. HBM is tied to the generational upgrades of AI GPUs. For each generation of GPU to increase token throughput, simply stacking compute power is not enough. If HBM capacity and bandwidth don't keep up, even the strongest GPU will be bottlenecked. If data isn't fed fast enough, utilization won't rise. Customers find the upgrade cost-performance ratio low, and NVIDIA's selling point is weakened. Therefore, HBM is no longer just an accessory, but part of the GPU generational upgrade. Moreover, several factors are superimposed: High technical barriers (3D stacking, packaged together with GPU), Deep customer validation (know that not just anyone who produces it can sell to NVIDIA), Few suppliers (three players: SK Hynix, Samsung, Micron), HBM usage per GPU is still increasing. Putting these things together, there is a new point of discussion: Will HBM transition from a pure cyclical product to a cyclical product with growth attributes? I think HBM's cyclical attributes remain, but the growth weight has increased. Of course, we shouldn't be too absolute; slowing cloud provider capital expenditure or unexpected over-expansion can both put pressure on prices. How do you view HBM's current pricing? Has the market already priced in too much growth premium?