Tether CEO looks to the future of AI: Efficient models reduce training costs, local and edge inference become a trend

On January 27th, Paolo Ardoino, CEO of Tether, tweeted that millions of GPUs will no longer be needed to train powerful models in the future. This pure reliance on brute-force computing is inefficient and inelegant. Better models will significantly reduce training costs, while access to data remains key. Inference will shift to local or edge end, avoiding wasting hundreds of billions of dollars on pure computing power.