Almost every chip company report this year featured three letters: HBM. Nova said its revenue in the field hit a record, Camtek sells systems that inspect HBM chips, and Qualcomm said it is raising prices partly because of the memory shortage. So what is it, and why does it matter so much?
HBM stands for High Bandwidth Memory. To see why it is needed, consider how an AI chip works. A chip like Nvidia's performs billions of calculations a second, but it has to get the data for them from memory. If memory is slow, the chip simply waits. In large AI models, which must move enormous amounts of data constantly, memory has become a bigger limit on speed than the processor itself.
HBM's solution is to build upward. Instead of one memory chip, several are stacked like floors in a building and connected by thousands of tiny vertical links. The stack is placed right next to the AI chip in the same 'package', so data travels a very short distance at very high speed. The result is much faster memory, but also much harder and more expensive to make.
That is where the shortage begins. Only three companies make HBM at meaningful scale: SK Hynix and Samsung of Korea and US-based Micron. When they shift lines to HBM, which pays better, less ordinary memory is left for phones, computers and consoles. That is one reason memory prices jumped this year and why phone and chip makers are raising prices. In practice, AI is making even unrelated devices more expensive.
And the Israeli link? Every floor in an HBM stack is another layer where something can go wrong, and every tiny link is a possible point of failure. So making HBM requires far more measurement and inspection than ordinary memory. That is exactly Nova and Camtek's field, which is why the global memory shortage translates into record quarters for them.
