Tim Davies shows compute costs drop roughly 10x every five years. By 2060, processing power equivalent to all human intelligence could cost $25 million instead of today's $500 billion in data center spending. That shift means AI moves to phones and laptops, automating work and reshaping which businesses win.
Questions this video answers
1936 to the back end of 2023.
A 20 quadrillion increase.
Every five years (≈10× reduction per five years).
A billion calculations per second.
$25 million.
Chapters
Ray Kurzweil has been analyzing the cost per computation per consistent dollar, the number of calculations that you can run per second for a consistent amount of money since 1936. And from 1936 to the back end of 2023, that was a logarithmic straight line representing a 20 quadrillion increase in the number of calculations that you can do per second. Now, I don't know about you, but 20, quadrillion is a big number, and I can't visualize it. But if we change the way that that graph is shown. So instead of being a big line going up showing the number of calculations that can be done per second, and instead we plot it as the cost for a certain amount of calculations, and then show that over time, we can look at the cost of carrying out a gigaflop, which is a billion calculations per second. We can see that there's a trend that every five years it reduces by an order of magnitude. What that means is effectively divide the number by 10 every five years, which is impressive. But when you plot on that same graph a human level of intelligence, you can see that very quickly that cost of intelligence, the cost of using the same amount of processing power as a human has within their entire brain, which we don't really apply to much very often, is coming down very quickly. And we've done a graph which looks at, okay, what about the cost of the intelligence of all of humanity? And if that's coming down on that same Trend line, by 2060, the cost of all of the intelligence of all of humanity would cost $25 million. That sounds like a lot of money, but given that there is 500 billion being invested in data centers right now, it's really not that much money. And so this trend is going to mean that more and more things that are currently done by humans will be done by artificial intelligence sat within data centers or by artificial intelligence sat locally on your phone. We're already running models on our laptops which are capable of answering complex questions. And so without the need of any Internet connection, you just using the processor on a laptop or a processor on a phone, you'll be able to run very, very comprehensive interactions with AI without the need for a data center. And that capability keeps on increasing. So what impact does that have within a business which employs hundreds of workers who are looking at problems, producing presentations, sharing those presentations by email? Every single one of those actions can be automated. And so within a business, if you're not thinking, what are the fundamental steps that we have where we take action where we're spending time. And you're looking to reduce the amount of time that's spent on those actions by using artificial intelligence to accelerate the process, you're being left behind. And businesses that are doing that well will be able to significantly out compete you and because they will have a lower cost of operating. So this will have an impact on the jobs market, but much quicker. It's going to have an impact on the competitive or one competitiveness of one business versus another. Whether you can compete on price will depend how much you're using technology.
