Jensen's basically saying stop obsessing over whether Microsoft or Meta can justify their capex bills. The compute buildout isn't a two-horse race — it's structural, it's multi-tenant, and the spending curve only goes one direction.
This matters because the bear case on $NVDA always circles back to "hyperscaler ROI anxiety." But if you zoom out, the capex isn't speculative anymore. It's infrastructure. The same way you didn't question fiber buildouts in the 2000s once streaming became obvious, you don't question GPU farms once inference economics pencil.
The tell here: Jensen's confident enough to redirect attention away from the marquee logos. That's not evasion — that's someone who knows the demand base is widening, not narrowing. Sovereign clouds, enterprise private clusters, regional SaaS players, even defense and biotech are spinning up their own stacks. The hyperscalers were the beachhead. Now it's a land grab across every vertical that can't afford to rent inference forever.
Bottom line — if compute spend is "only going higher," then the supply chain under it (memory bandwidth, CoWoS capacity, optical interconnects, power infrastructure) stays tight for years. That's the thesis. And it's not about one earnings call — it's about a multi-year capex super-cycle that hasn't even hit enterprise peak yet.
This matters because the bear case on $NVDA always circles back to "hyperscaler ROI anxiety." But if you zoom out, the capex isn't speculative anymore. It's infrastructure. The same way you didn't question fiber buildouts in the 2000s once streaming became obvious, you don't question GPU farms once inference economics pencil.
The tell here: Jensen's confident enough to redirect attention away from the marquee logos. That's not evasion — that's someone who knows the demand base is widening, not narrowing. Sovereign clouds, enterprise private clusters, regional SaaS players, even defense and biotech are spinning up their own stacks. The hyperscalers were the beachhead. Now it's a land grab across every vertical that can't afford to rent inference forever.
Bottom line — if compute spend is "only going higher," then the supply chain under it (memory bandwidth, CoWoS capacity, optical interconnects, power infrastructure) stays tight for years. That's the thesis. And it's not about one earnings call — it's about a multi-year capex super-cycle that hasn't even hit enterprise peak yet.