In the many conversations I’ve had with business leaders and policy makers, there’s no doubt that the seemingly huge sums AI companies spend on chips and power purchase agreements are to blame, and the real cost of AI is increasing. continues to do so. Nvidia’s top customers spent $4.2 billion on Nvidia chips and services in the last fiscal quarter, while Microsoft signed a record purchasing power deal with Brookfield worth an estimated $10 billion.
These staggering numbers are the result of land grabs to build AI platforms. Demis Hassabis, head of DeepMind, said Google will spend more than $100 billion on AI over the long term. “The risk of underinvestment is dramatically greater than the risk of overinvestment here,” said Sundar Pichai, Mr. Demis’ boss. Many people think this is a waste of money. Or will an unsustainable loss leader rebound sharply?
Perhaps this money is being wasted, but I don’t know what the problem is. If big tech companies are “wasting money”, do they need our sympathy? Investors can sell their shares and move elsewhere. If a manager fails, he or she can be fired. life goes on.
But anyway, that’s not what’s happening.
The cost of delivering AI services (whether LLM or lower) at a given quality is rapidly decreasing. There are two reasons for doing so:
a. Exponential Trends in Processing – Fan’s Law is the new Moore’s Law, but faster. Every year, Nvidia expects to deliver chips that are two to three times cheaper than the previous year with the same performance. This is more than twice the rhythm of Moore’s Law.
