Sometimes They Really Do Ring a Bell, Part II
Back in October, I asked if Oracle (ORCL) was the canary in the coalmine signaling trouble in AI land. The company announced at the time that it’s AI-related business wasn’t all that profitable.
The margins were a rather unimpressive 14%. Oracle’s legacy software and services businesses generate margins of around 70%.
It turns out that spending billions open billions of dollars on infrastructure that depreciates rapidly and requires boatloads of money to maintain might not be a great business model. (See "The Achilles Heel of the AI Boom” for an excellent explanation from my friend Kai Wu on how excessive capital spending is turning America’s finest tech companies into bloated utilities.)
Well, Oracle is back in the news this week.
The Financial Times reported this week that Oracle’s largest data-center partner, Blue Owl Capital, had backed away from a planned $10 billion data-center project.
There are plenty of innocent explanations. Maybe Blue Owl wanted better terms and decided to play hardball. Maybe Oracle got cocky and overreached on pricing. Maybe the Blue Owl CEO simply had a terrible round of golf and wasn’t in the mood to cut a deal.
Or maybe…
Maybe Blue Owl looked at the sheer volume of data-center projects coming online and got cold feet. Maybe they’re starting to question whether the AI profit model has really been proven out — and whether today’s infrastructure spending boom is setting us up for a colossal bust and a massive supply glut of compute.
We’ve seen this movie before, of course.
Infrastructure Busts Create Massive Opportunities
Remember Global Crossing?
I’m betting you probably don’t. The company went bankrupt back in 2002… one of the largest bankruptcies in history at the time. But in its short life, the company made a massive impact.
In 1997, former investment banker Gary Winnick led a group of entrepreneurs to launch Global Crossing. Winnick saw the massive potential of the internet and realized immediately that the biggest impediment to wider adoption was bandwidth. So, his mission was to crisscross the world with a massive global network of undersea fiber optic cables.
By 2001, it had connected 200 major cities in 27 countries and laid an estimated 100,000 miles of cable. That’s roughly enough to wrap around the earth four times… or get you a third of the way to the moon.
But it wasn’t the only company doing this.
Before the turn of the century, companies laid 80 to 90 million miles of fiber-optic cable. That’s enough to get you a quarter of the way to Mars!
We know what happened next. The overcapacity reached an extreme, and most of the cable remained unused “dark fiber.” Even four years after the bubble burst in 2000, an estimated 85% to 95% of the fiber laid in the 1990s was unused.
If the story ended here, it would be a tragic destruction of wealth. But that’s not where this story ends.
The massive oversupply of internet cable caused the price of communications to plummet. And that created new opportunities that no one could have dreamed of back in the 1990s.
Netflix (NFLX) might be ubiquitous today as the world’s leading streaming service. But that’s not how it started. It was founded in 1997 as a DVD by mail service.
Well, starting in 2007, Netflix took advantage of the dirt-cheap bandwidth following the tech crash to launch the streaming service we know and love and started a revolution in viewing habits that continues to this day.
It’s hard to imagine life without YouTube today. But YouTube, like Netflix, would have never been a viable business without the massive investment in internet bandwidth made by Global Crossing and others. An investment that Global Crossing itself never actually got to enjoy the benefits of.
Facebook? Instagram? Spotify?
Same thing. All of these companies were able to launch viable businesses because the cost of internet bandwidth had already dropped like a rock.
We’re going to see something similar happen this time around in AI.
Something in the ballpark of $400 billion has already been spent in 2025 on AI infrastructure like datacenters. Next year, the spending is expected to be around half a trillion dollars. It’s an amount of money so ridiculously large that is hard to fathom.
Eventually, we’ll get a glut. There will be more AI capacity than currently needed, and the price of compute will collapse.
Many of the companies building out the AI infrastructure today will likely take massive losses. Some will probably go bankrupt. That’s what happens after a bubble bursts.
But the collapse in price will create massive new opportunities for companies that haven’t even been founded yet. AI – much like the internet – will become something resembling a public utility. It will be everywhere… embedded in everything… and dirt cheap to use.
Look at the success of Big Data company Palantir (PLTR), one of the great success stories of the AI age. Now imagine dozens or even hundreds of similar success stories.
Great!
But what about now?
You Should Start Taking Profits If You Haven’t Already
I cannot say with certainty that the top is in place for the Nvidia (NVDA), the Mag 7, the tech sector as a whole or the entire S&P 500. In the short-term, the market is driven entirely by emotions and short-term money movements.
But I can tell you that the warning signs are in place. Mom and pop investors are actively seeking risk. There is no fear… or rather the only fear is the fear of missing out — FOMO. Stock valuations are at records by most traditional metrics. And thus far, the only companies making money in AI are the ones selling the infrastructure… which they are selling to each other in a closed loop.
You don’t have to sell everything and hide in a bunker in Idaho. Please don’t do that.
But you absolutely should be taking risk off the table right now. Rebalance your portfolio, at least shaving off a little bit of the tech exposure. Have some exposure to income and value stocks, many of which are trading at really attractive prices today.
And have a little cash on hand.
I don’t know you and can’t tell you how to invest your portfolio in a Substack post. But I can tell you this…
If a “normal” portfolio for you is something like a 60/40 portfolio, you should consider dropping it to something closer to a 50/50 or a 40/60. If you’re normally more heavily invested, say, 80/20, maybe dial it back to a 70/30 or 60/40.
(If you want something more personalized, please contact me. I’m happy to sit down with you and go over your portfolio. I’ve done this for hundreds of clients over the years, and I’m happy to do the same for you.)
You’ll still have plenty of skin in the game if turns out that I’m wrong and AI ends up being an even bigger profit windfall than the market is pricing in. But if I’m right, and we really are close to a colossal bust, you’ll be glad you weren’t fully invested.
Charles Sizemore, CFA