Maybe AI Is Finally Starting To Pay The Tab
AI is expensive. Like, really expensive.
We get it...we've been beat over the head about how hundreds of billions of dollars are being poured into data centers, chips, power generation, transmission, cooling systems and whatever else. Just so Claude can tell me how to cook a steak without disappointing my wife (it's a learning process...).
And for the last few years, investors have mostly focused on that side of the equation.
How much are they spending?
Where is all this money coming from?
Are they ever going to earn an adequate return on it?
Hell, those are great questions that do not have easy, straight forward answers.
I wrote recently about one consequence of this spending boom that I think is being under-appreciated: AI may be creating such an enormous demand for capital that it's actually contributing to higher long-term interest rates.
But there's another side of this equation.
If we're going to spend this much time talking about what AI costs, eventually we should probably ask what we're getting for it. You know - like what we should probably ask the U.S. government when they make a policy decision...any decision for that matter.
Anyway...and we're starting to get some interesting answers.
Productivity is one of those economic statistics that sounds unbelievably boring until you realize it's basically the closest thing economics has to a cheat code.
If I hire another employee and produce more stuff, cool.
If I build another factory and produce more stuff, also cool.
But if the same employee, working the same number of hours, can suddenly produce more stuff?
Now we're cooking. That's productivity growth.
And lately, productivity has been looking pretty darn good.
In the latest data from the BLS, non-financial corporate productivity increased 3.1% over the past year. Manufacturing productivity increased as well, while manufacturing unit labor costs actually declined.
In plain English: companies are finding ways to produce more without their labor costs rising at the same rate.
Now...Is that AI? I have absolutely no idea.
And neither do any of the "experts" on TV who confidently tell you it is. Don't believe me? Go ahead and build the polynomial. Start with interest rates, fiscal spending, capital investment, labor markets, demographics, energy prices, supply chains, regulation, technology, business formation, and a few hundred other variables.
Then figure out how each one interacts with all the others. Oh, and the coefficients change over time.
Good luck!
A $30+ trillion economy isn't exactly a controlled science experiment.
But here's what we can say: For three years, we've been told that AI would eventually make workers dramatically more productive.
So if that "thesis" is right, shouldn't we at least start looking for clues in the productivity data?
Because we're beginning to see some.
And then Accenture reported earnings.
After digging around to try and find those said clues, I found a pretty decent one.
Out of all the industries out there you would think that consulting would be one of the first to get hit by the AI train.
Why pay a human to spend hours researching something, analyzing data, writing code or creating a PowerPoint when increasingly capable AI models can do portions of those tasks in seconds?
Not exactly the greatest sales pitch for hiring another consultant.
Except Accenture's latest results didn't look like a company waiting around to be replaced.
The company specifically discussed AI's role in improving how work gets done, including higher revenue per employee.
And that is where I think the AI story starts getting much more fun and less doom and gloom.
We've mostly been presented with two possible futures.
In one, AI replaces half the workforce and we're all unemployed by Tuesday.
In the other, Big Tech spends trillions of dollars building enormous data centers that never generate enough revenue to justify the investment.
But there is a third possibility that gets considerably less attention.
What if AI mostly makes existing businesses better?
Not overnight.
Not everywhere.
And certainly not without some jobs and industries getting disrupted along the way.
But imagine a company where 10 people can eventually accomplish what previously required 20.
Or a software developer can write 100% more code.
Or an analyst spends 30 minutes gathering information instead of eight hours.
Multiply relatively small improvements like that across millions of workers and thousands of companies and suddenly we're talking about something much bigger than selling $20/month chatbot subscriptions.
We're talking about margins and ultimately, profits. Check out the surge in S&P 500 EPS within the last few quarters (yellow highlighted area).
Corporate earnings growth has been exceptionally strong recently.
Again, I'm not saying AI is responsible for that. It isn't.
But this is why I'm increasingly interested in watching productivity alongside all of those gigantic AI capex numbers.
The debate over the last few years has largely been: Can these companies possibly make enough money from AI to justify what they're spending?
And I'm not saying it's the wrong question, but maybe we can do one better.
Maybe the return doesn't only show up as a new line item labeled "AI Revenue".
Maybe some of the return shows up in a bank processing loans faster (maybe all of that shifts to the blockchain???).
Or an insurance company handling claims with fewer hours of human work.
Or a manufacturer reducing downtime.
Or a consulting company generating more revenue per employee.
That's much harder to put into a Wall-Street style investor presentation, but economically, it could be far more important.
And this doesn't mean the AI spending boom automatically works out (but AI is definitely here to stay whether you like it or not so buckle up).
At today's valuations, companies still need to deliver. We're talking about an extraordinary amount of capital being deployed, and history has plenty of examples of transformative technologies that were very real...and investments in them that were very bad.
The internet did, in fact, change the world. That didn't make every dot-com stock a good idea.
But that's exactly why we're watching this.
The question is no longer whether companies are spending an absurd amount of money on AI.
They obviously are.
Now we get to watch whether all that spending actually makes the economy more productive and corporate America more profitable.
And for the first time, we might be starting to see some receipts.
