Zuckerberg just delivered one of the sharpest rebuttals to the GPU depreciation bear thesis.
He framed owning data centers as a financial advantage for $META — not a burden. The company will likely keep running GPUs well past their accounting depreciation window.
Even more telling: he said an $NVDA H100, a chip that's years old now, is worth MORE today than it was last year.
The bear case has always leaned on the idea that GPUs lose value fast and crush AI margins within a couple years. But if older chips remain in demand and continue doing productive work, the payoff window is much longer than the market prices in.
This isn't just good for $META's unit economics. It's a quiet validation of $NVDA's installed base holding value longer than expected.
When depreciation schedules don't match real-world utility, that gap creates opportunity.
He framed owning data centers as a financial advantage for $META — not a burden. The company will likely keep running GPUs well past their accounting depreciation window.
Even more telling: he said an $NVDA H100, a chip that's years old now, is worth MORE today than it was last year.
The bear case has always leaned on the idea that GPUs lose value fast and crush AI margins within a couple years. But if older chips remain in demand and continue doing productive work, the payoff window is much longer than the market prices in.
This isn't just good for $META's unit economics. It's a quiet validation of $NVDA's installed base holding value longer than expected.
When depreciation schedules don't match real-world utility, that gap creates opportunity.