Nvidia is pushing its new “Vera Rubin” chip platform with claims that AI costs could fall by up to 90%.
That idea is doing more work than the chip itself. It is helping hold up expectations that AI will keep getting dramatically cheaper even while spending on AI infrastructure keeps rising.
The problem is simple. If AI does not get much cheaper as fast as expected, the current level of investment starts to look very different. Not like a smooth growth story, but like a very expensive bet on future efficiency.
Right now the market is treating cost reduction as if it is already guaranteed. That is the part holding everything together.
The deeper shift is this. AI companies are no longer being judged mainly on what they build today, but on how cheaply they are expected to run in the future. That future assumption is now carrying present valuations.
If that assumption slows down or fails to show up in real numbers, the pricing behind the whole AI buildout has to adjust.
That is the real story underneath the chip announcement.