# chip earnings — X 热门讨论 (2026-09-30 19:42 UTC)
## @ren_stocks (Ren) · 09-30 17:21 · ♥31 ↻0 💬18 $NVDA $NBIS
The biggest bear case on AI infrastructure has been that GPUs lose their value in two or three years, so everyone building data centers is overstating their earnings. This chart puts real prices on that argument.
Using Silicon Data's pricing, a six-year-old A100 still holds about 25% of its purchase price, which is where an accelerated depreciation schedule expects it to be at year two. An H100, four years after mass availability, is still worth 58%. And a B200 is trading at 158% of what it cost, a sign of how far demand is running ahead of supply.
That changes the math for everyone who owns compute. If a GPU is still worth more than half its price after four years, a five or six-year depreciation schedule starts to look conservative.
The neoclouds benefit the most, because GPUs are their collateral, their balance sheet and their product all at once, so every point of residual value makes financing cheaper and returns higher.
The honest caveat: it's Nvidia's own chart, the purchase prices are third-party estimates, and a B200 above 100% is a shortage price that won't last forever. But even the oldest chip on the chart is beating the bear case by four years.
Old compute keeps getting put back to work, and that's one of the strongest arguments for the whole buildout.
We are crushing the bears over here ⚡️ https://x.com/ren_stocks/status/2105347306421916129