NVIDIA shared a video of Jensen Huang’s speech from the Cadence Live 2026 event on the X platform. During his address, CEO Jensen Huang underscored that the full-stack approach is the linchpin of NVIDIA’s dominance in the AI arena and introduced the concept of “generating tokens at the world’s lowest cost.” He highlighted that while NVIDIA’s AI hardware comes with a hefty price tag, by fine-tuning the software stack, it can unlock the hardware’s full potential and attain the world’s most economical token generation costs. The velocity and expenditure associated with token generation are pivotal in gauging the efficiency and worth of AI systems. Huang opined that one shouldn’t solely depend on raw hardware computational might but should harness software guidance to boost hardware efficiency to its peak. NVIDIA’s CUDA ecosystem acts as the pivotal link between hardware computational prowess and software applications. When queried (which can be interpreted as facing “criticism” or “inquiries”) about the steep hardware prices, he elucidated that by spreading the hardware costs across vast output, NVIDIA secures the lowest cost per token. Moreover, the system’s energy efficiency ratio remains low, aiding in curbing operational outlays. He further suggested that the primary yardstick for assessing the value of AI systems ought to be the cost per token.