Many people worry that storage will end up like the fiber optics of 2000—because overbuilding leads to excess capacity, and ultimately something breaks. But I think storage and fiber optics still differ in a few fundamental physical aspects:

1) The “linearity” of Maxwell’s equations means that a single optical path can be reused across multiple frequencies without interference; therefore, the bandwidth of a single fiber can be effectively infinite.
2) Even if you use the same frequency, the fiber can still improve communication efficiency through purely mathematical methods like modulation and coding.
But the problem with HBM right now has nothing to do with “efficiency”—it’s about capacity. A Rubin GPU with 500GB of video memory needs twice as many HBM chips as a 250GB Blackwell GPU. That’s an unchangeable mathematical fact.
In addition, for every extra bit in video memory, you need additional physical space, which also wastes a bit more wafer. This is unlike what exists in the virtual world, where—so long as energy conservation is satisfied—you can create photons more or less arbitrarily, without limit.
Therefore, likening storage to fiber optics isn’t appropriate. Even if there is “excess,” the difficulty and the timeline will be far stronger than—and come far later than—the internet infrastructure bubble of 2000.
Even if researchers develop storage-specific “byte reuse” or context-compression algorithms to reduce demand, that again runs into Jevons’ paradox.
The higher the storage efficiency, the more people use it—so even if total demand remains the same, it doesn’t stay put; total demand increases.
It’s hard!