This is the #1 question investors are asking companies making big AI investments right now.
While major technology companies are expected to spend approximately $750B on artificial intelligence infrastructure this year, they are also expected to earn $175 billion from artificial intelligence revenues.
This means a deficit of $575B.
There is a big problem with this comparison…
A large AI data center takes approximately two years to build and install. The owner must provide land, provide electricity, install cooling systems, and assemble thousands of servers.
Money starts flowing out long before the first customer logs in.
Therefore, most of the artificial intelligence revenues emerging today come from equipment purchased in 2023 and 2024. Most of the $750B spent this year will not earn a penny until 2028.
Comparing this year's expenses with this year's income is telling the truth backwards. Companies like $GOOGL and $AMZN are also using their new chips to train models first. And this education does not directly earn them a penny.
But then these chips switch to “inference” mode, which is a fancy term used to describe the process where AI uses what it has previously learned to give you an answer.
That's when the safe starts ringing.
Does a chip pay for itself?
Nvidia's entry-level A100 server launched in 2020 at $199,000. There are eight chips in the box. If you run it at full capacity for five years, each chip costs about 65 cents per hour (including electricity).
How much per hour is the same six-year-old chip being rented today? For about $1.64. If you rent from Amazon it costs $3.72 per hour.
In the computer world, old hardware often becomes scrap. However, artificial intelligence chips age just like fine wine. Even Nvidia's four-year-old H100 chip has gained value. Rental prices rose nearly 40% in March after bottoming out last October.
$AMZN announced that its AWS cloud business grew by 37% last quarter…
It is the fastest growth rate recorded in more than four years.
$MSFT reported a 43% increase in its Azure cloud service. Google's cloud revenues increased by 82%.
These cloud platforms are almost two decades old. As large businesses grow, they normally slow down. Artificial intelligence has given new momentum to their sails.
Artificial intelligence is also elevating businesses that already earn billions of dollars. Google's search revenues rose 17% last quarter. $META advertising revenues jumped 27%.
Artificial intelligence is helping Google answer tougher questions. It helps Meta show people ads they will click on. These gains are included in the current income items. That's why they don't appear regularly under "AI Revenue."
The most interesting evidence comes from a shipping company with a century of history: CH Robinson Worldwide (CHRW) now quotes shipping in 32 seconds. This process used to take 1,545 minutes (about 25 hours).
Three-quarters of small shipping orders are handled completely without human intervention. In this way, the company saves approximately 600 hours of labor per day.
Operating expenses fell 12.6% in one year; This is a huge percentage for a low profit margin business. But this doesn't show up anywhere in AI "return on investment" calculations.
If AI processing power were increased excessively, prices would fall.
They're doing just the opposite.
In January, Amazon raised the price of AI cloud capacity by nearly 15%. This was the first increase in a period when cloud prices had only been declining in almost 20 years. The company raised prices again on July 1.
Listen to $ORCL CEO Larry Ellison describe his dinner with Elon Musk and Nvidia CEO Jensen Huang:
This is how I would describe dinner: Oracle, me, and Elon were begging Jensen for GPUs. Please take our money. No, no, get more. You're not getting enough.
Once upon a time, pioneering AI labs were seen as bottomless pits of money. But did you know that Anthropic, the company Claude produces, is now cash flow positive?
It started this year with a $9B annual revenue rate. By May, it exceeded $47B, growing 5 times in just four months!
Anthropic also anticipates making its first operating profit this quarter.
Demand is so high that it has to turn away customers. As demand consumed processing power faster than it could buy it, it set usage limits and cut off relationships with partners.
There is a risk that I am aware of…
The five largest AI companies doubled their debt in five years, adding about $350 billion in total debt. Alphabet had a negative cash flow quarter for the first time since its IPO in 2004.
The real danger here is excessive debt.
But there is a big difference between the current situation and the bursting of the dot-com bubble. At that time, telecom companies laid fiber optic cables, and 85% of them remained unused for a decade.
Nowadays, equipment runs out before the concrete hardens.
Follow the $750B project…
While big tech giants are getting a return on investment from their AI infrastructure, they are still spending huge amounts of money.
That's why we choose to invest in companies that benefit from this flood of spending, rather than companies that spend money. In short, invest in companies that pay off…
Nvidia sells leading AI chips and is trading at its lowest price-to-earnings ratio in years.
$TSM
produces the most advanced chips. ́
$MU
provides the high bandwidth memory required by these chips.
Many smaller players also benefit from the same source; these include power producers, cooling specialists, electrical equipment manufacturers and data center component suppliers.
