The Bullish Bear
When the rational move is not to play the game
Frankly, I’m sick of talking about SpaceX. And you might be sick of hearing about it.
No, I wasn’t chasing an IPO allocation, nor am I considering buying the stock any time soon.
Does that make me a bear on the company?
Absolutely not!
I’m also not chasing the AI infrastructure plays. Not chip stocks, or memory players, or data center buildout guys.
Does that make me bearish on AI?
No way!
In fact, quite the opposite. It’s because I’m bullish on the technology that I feel like the companies are a bad deal at their current prices.
Let me explain.
Disclaimer: I’m no technology expert. These are just the ramblings of a generalist, and might be way off base.
The story of progress
Let’s go back to the dawn of modern computing, and compare it to where we are today. I asked my trusty AI chatbot to help me model the amount of physical space and power consumption necessary to match the power of the current iPhone 17 Pro.
In 1970, it would have taken about 69,000 CDC Supercomputers, consuming 6.6 gigawatts of energy. The space required to hold this equipment? About 156 acres.
Fast forward to 1980, and we can now use 15,500 Cray-1 Supercomputers. Energy use and real estate have both fallen dramatically, down to 1.8 gigawatts and 25 acres, respectively.
Now we are in 1990, and Moore’s Law continues to work its magic. Your iPhone is now equivalent to 930 Cray Y-MP8s, consuming a modest 112 megawatts and requiring only 1 acre of computer cabinets. The kind of thing that can fit inside a large well-funded research facility.
Improvement in the 1990s was dramatic, and by 2000 the IBM ASCI White was the world’s fastest supercomputer, about 4x as powerful as your iPhone.
By 2010, a high end PC with 2 Nvidia graphics cards was now able to match the slim thing you now carry with you everywhere in your jeans pocket.
Where are we now?
What’s the point of this history lesson?
If we see even a fraction of this type of progress in AI compute, some of today’s infrastructure may prove far less valuable than investors currently assume. And if we won’t need these data centers here on Earth, we almost certainly won’t need them in freaking space.
The buildout is happening because technology is moving fast, and because the companies investing in it are afraid of being left behind over the next few years. They really have no choice but to participate in this high-stakes prisoners’ dilemma. But the end result of this gold rush is almost certainly a vast overinvestment in chips and infrastructure that will likely go unused, or need to be repurposed. Consequently, the dream of data centers in space may be much further away than the current enthusiasm implies… if we see it at all.
An investment in these companies at today’s prices requires us to believe that demand growth will outrun hardware improvement, software efficiency, and the fruits of competition. It is, in my opinion, a bearish view on technological progress and human ingenuity.
Progress isn’t anywhere close to being done. The next Elon Musk is waiting in the wings, and he or she will drive innovations that make a data center the size of “a significant part of the footprint of Manhattan” look like a huge waste of capital.
The current AI infrastructure buildout is rational, and will form the basis for lower cost compute and the increased demand we will certainly see in the future. But that doesn’t mean it will be a good investment for the companies making it. It’s the difficulty of making this call that leads me to avoid making any big bets on the future we all see coming.
The other side of this argument is that as prices decline, demand will surge. And indeed it will. We are still only in the very early stages of witnessing the impact of AI on our lives. But before you simply extrapolate recent growth into the future, consider what happened in the 1990’s telecom boom.
What does history tell us?
The 1990s analogue is the growth in bandwidth use.
Every bullish prediction back then was correct. In fact, the growth in bandwidth use was probably underestimated.
From 2000 to 2025, global internet traffic grew from roughly 0.9 exabytes/year to about 8,800 exabytes/year. That is roughly 9,000x to 10,000x growth, equivalent to an annual growth rate of 44% per year for 25 years.
Today, we stream almost everything we watch over the internet, often in 4K video. We hold videoconferences with hundreds of simultaneous video participants. We stream video games with almost zero latency. All while endlessly scrolling on our handheld devices.
We have seen insane bandwidth demand growth, and yet even with that growth, many of the telecom infrastructure providers who literally laid the groundwork for the internet of today disappeared, or took decades to recover the former highs. The vision was realized, but they were not good investments. And it is with that hindsight that I am happy to bow out of participating in the current mania. Those who choose to do so should be aware of the history.
If you weren’t paying attention back then, take a look at this Wired magazine article from 2004: Bandwidth Glut Lives On.
"The bandwidth supplied from the late 1990s to about 2001 or so grew exponentially," he said. "Only a small fraction of it was lit, but it was still far, far more than the market would absorb.
There’s no reason to believe that the story will be different this time around. AI may change the world. SpaceX may remain one of the most remarkable companies ever built. Data centers may become essential infrastructure for the next decade. All of that can be true, and today’s prices can still be wrong.
Jockeying for position in a rapidly changing market can create massive winners, and massive losers. Sometimes they can even be the same company.
Bottom line: A great investment requires a great price, not just a great story.