Part 3/11:
The most innovative idea presented is converting electricity into AI tokens for export. Unlike physical commodities like oil, digital tokens can be transmitted at nearly zero marginal cost, traveling at the speed of light. This enables surplus energy to be transformed into computational resources, which can then be sold globally, effectively turning energy into a new kind of export akin to traditional oil.
This strategy leverages the key distinction between model training and inference:
Training large models is resource-intensive and happens infrequently.
Inference, the real-time querying and utilization of models, is continuous and scales exponentially with usage.