AI will only profoundly change the world - and unlock $15 trillion of economic value over the next 10 years - if it is cheap enough to be deployed abundantly. 

That brings us to a discussion of the second factor which will drive forward the AI Revolution of the 2020s: costs. 

AI is built on top of computers, which benefit from both Moore's Law - the number of transistors on a semiconductor chip will double every two years - and Wright's Law - the cost of production decreases at a constant percentage for every cumulative doubling of production volume. Therefore, the computing costs for AI - which have been plunging for years - should continue to drop for the near future. 

Not to mention, as data becomes more abundant in the coming years, the cost of data itself will drop, too. And so will the costs to gather, analyze, and store it. 

Put it all together, and AI training costs should crash in the 2020s. 

The cost to train a large language model to GPT-3 level performance was $4.6 million in 2020. Last year, it dropped to $450,000 - below even aggressive forecasts from ARK Invest for a drop to $740,000. 

If this cost-decline curve persists, AI training costs could fall to $30 per model by 2030 - down 99.999% from 2020.