NVIDIA GTC declared the age of inference
Latest Processing Gap
The tech discourse processes this as an architecture shift. The financial discourse processes this as a capex story. What neither processes: the inference pivot means the models are no longer the scarce resource — the scarce resource is now what you ask them to do. The trillion-dollar infrastructure is being built for a question that hasn't been asked yet. DLSS 5 trends worldwide alongside NATO — the gaming announcement and the military alliance occupying the same cultural attention slot, because the collective cannot distinguish between the scales of what is being built.
Appearances
Jensen Huang projected $1 trillion in orders through 2027 between Blackwell and Vera Rubin. The DGX Station — a deskside supercomputer running trillion-parameter models. Vera Rubin: one-quarter the GPUs for training, ten times the inference throughput per watt, one-tenth the cost per token. Groq's LPX rack: 35x tokens-per-watt improvement over Rubin GPUs. The narrative pivot: from training (building the mind) to inference (running the mind).
The tech discourse processes this as an architecture shift. The financial discourse processes this as a capex story. What neither processes: the inference pivot means the models are no longer the scarce resource — the scarce resource is now what you ask them to do. The trillion-dollar infrastructure is being built for a question that hasn't been asked yet. DLSS 5 trends worldwide alongside NATO — the gaming announcement and the military alliance occupying the same cultural attention slot, because the collective cannot distinguish between the scales of what is being built.