AI Pays Off Four Ways
Disruptive technologies like AI pay off in the wider economy four times as they spread.
First, through a faster rate of investment. Big tech companies that used to sit on cash now spend it to build data centers. Those data centers hire people to dig up ground and install equipment. The return on that invested capital is much higher than in other uses. The market is badly under-forecasting the return on invested capital in infrastructure-as-a-service right now, shown in CoreWeave's (CRWV) quarter.
Second, higher yields from that faster return.
Third, by creating market activity that did not exist before. Driving is the clear case. Robotaxis are about to drive people around. Today 99% of driving miles are done by hand, a waste of human brains. People will pay robotaxi services to hand off that task.
Fourth, freeing people to do other things - watching Netflix (NFLX) or TikTok, or starting a new business.
Today's data center investment is only the first step of this four-step process that speeds up the growth rate of the whole economy.
The Statistic Already Moving
The change already shows up in capital formation and capex. At the company level, firms that use AI more heavily are hiring more workers. The idea that AI will cut jobs is wrong; the opposite is happening. Firms using AI find more ways to deliver services to customers, so they need more people to run these tools well.
Real GDP growth will top 7% a year by 2030. That would be a jump in growth not seen since electrification, the telephone, and the internal combustion engine arrived around 1900. This is a change in economic activity that plays out over this decade. The timing suits markets well, because people doubt that even the next data center will earn a return - a strong setup for equity investors.
The World Is Short Compute
The world is massively short on compute. Far more is needed. People doubt this cycle, but penetration is very low. A little over 20% of smartphone users use AI chatbots like ChatGPT just to ask questions. Only 2 to 5% of knowledge workers actually use agentic tools like Codex and Claude Code.
That low-20s penetration matches internet penetration in 1996 - two years before Google was even founded. The 1 to 5% level matches 1990, during the early internet boom, before most people knew what a browser was. This cycle is very early.
As more users come on and knowledge workers learn to run agents for themselves, compute demand will scream higher - trillions of dollars in demand by 2030. That drives demand for Nvidia's (NVDA) chips, AMD's (AMD) chips, AWS (AMZN) services, and CoreWeave's (CRWV) services. People are badly underestimating the scale of what is coming. The firm bought more Nvidia (NVDA) on a brief dip in its shares. Chips are becoming a new asset class to invest in, with new check writers now in the market.
Deflationary or Inflationary?
Question: Will AI and AI productivity end up deflationary or inflationary?
In knowledge work, people use AI to do more. In the embodied space - robotaxis and humanoid robots - costs of certain tasks will collapse, such as getting around town. At maturity, the cash cost of running a robotaxi service will be lower than the marginal cost of driving a car you already own and fully paid off. People will pay more than that because they save time, but the cost of getting from place to place still collapses.
Lower cost does not cut end demand. When the cost of things falls, especially in technology, people use far more of them. A parent who will not drive a kid across town to a movie or a friend's house will happily put the kid in a cheap-per-mile robotaxi. Falling costs set off an explosion in economic activity tied to them.
