Our TCO is So Good That Even When the Competitor's Chips are Free It's Not Cheap Enough

In the last decade, we have reduced the cost of computing by 1 million times and the cost of deep learning by the same amount. If we could reduce the marginal cost of computing to nearly zero, we might achieve something extraordinary, such as using large language models to extract all digital human knowledge from the internet. This concept is only feasible if the marginal cost of computing is zero. We've made that breakthrough, enabling a new way of doing software.

Accelerated Computing, which took three decades to develop, is likely the single greatest invention in the technology industry and could be the most important development of the 21st century. The H100 GPU chip, which weighs 70 lbs and consists of 35,000 parts, replaces an entire data center of old CPUs. Despite its high cost of a quarter million dollars per chip, the savings it provides are incredible.

In the next decade, we will increase computational capability for deep learning by another million times. This will lead to continuous learning, where the training and inference processes become one. The computer will continuously learn, infer, and be grounded by real-world data and synthetic data generated in real-time.

When the marginal cost of computing is driven down to zero, many new possibilities open up. Just like the reduced cost of transportation allows for more travel, reduced computing costs will enable more computation. This shift will transform the way we approach AI and computing, making it more accessible and efficient.

Nvidia's accelerated computing platform is the only one that is truly ubiquitous, allowing applications to run everywhere. Despite competition, Nvidia's focus on total cost of operations (TCO) ensures that even when competitors offer free chips, they cannot match Nvidia's value.

The original article: https://m.youtube.com/watch?v=oNwoA5akBlg