Notes from Camp Kotok: The AI Boom Is Just Getting Started
One thing that won't happen, however, is data centers in space.
Hello, friends. I’m in northern Maine at Camp Kotok, a gathering of about 50 financiers, economists, entrepreneurs, and otherwise interesting people, organized by David Kotok of Cumberland Advisors. This is the 25th straight year the event has taken place. It’s my sixth time here. This is the view from the deck at Leen’s Lodge, on West Grand Lake, where attendees argue about monetary policy and the intricacies of the bond market.
This is an informal gathering where discussion is off-the-record unless stipulated otherwise. I’m able to share observations.
Artificial intelligence is a main topic of conversation. We’re discussing the obvious questions: Is the remarkable AI-driven stock-market rally sustainable? What could slow it down? Where are we in the AI buildout? What could go wrong?
There seems to be a broad view that this revolution is just getting started. One person who likens it to a baseball game says we’re at the singing of the national anthem—not even in the first inning yet. Another points out that the global semiconductor industry is still relatively small, valued at less than $1 trillion. The global new vehicle market, as a comparison, is about $2.5 trillion, and the auto industry as a whole is a $4.5 trillion market if you include parts and service.
[Check out The Weekly Matterboard, where we rate the importance of the latest news]
Somebody pointed out that former Federal Reserve Chair Alan Greenspan uttered his famous remark about “irrational exuberance” in 1996. He was talking about a stock market that seemed overinflated. During the next three-and-a-half years, the S&P 500 rose another 105%. Don’t take that literally to mean we’re at the “irrational exuberance” moment of the AI boom and we’ve got another three years to party. It’s just context.
Nobody knows if the hyperscalers such as Amazon, Microsoft and Alphabet, which are spending hundreds of billions of dollars to build AI data centers and all other types of infrastructure, are going to recoup their investments. It could be years before clues emerge.
Is there an analog to the dot-com boom, which became the dot-com bust in 2000 and took 49% off the S&P 500 during a 31-month period? One difference is that many of the companies spending billions today are hugely profitable, rather than speculative startups. But the manic data-center buildout could be similar to the massive overbuild of fiber-optic networks in the late 1990s.
[See why the gigantic national debt is finally starting to matter]
At some point, data-center construction will slow, which will be a headwind for physical-infrastructure stocks such as Caterpillar, which has soared 102% during the last year. The construction aspect of AI probably will turn out to be cyclical, with slower periods when the buildout takes a breather. People refer to this as a possible “tools downturn.”
It’s an open question whether we’ll end up with a glut of data centers, with some going unused. The demand for computing power, or “compute,” in AI-speak, is so strong that there’s no sign of overcapacity, for now.
What will happen is new data centers with leading-edge chips will continually replace older data centers with lagging-edge chips. That doesn’t mean older data centers will become obsolete. There’s a lot of demand for lagging-edge chips that can handle routine tasks, or tasks that will become routine. Old chips don’t become obsolete just because new chips are better. There’s still important work they can do.
The chip industry is beginning to focus a lot more on cooling and energy efficiency, to address the problem of all the power needed to run data centers and the water needed to cool them. This brings us to Elon Musk’s bold idea to run data centers in space, where they could be powered by the sun and cooled by frigid ambient temperatures.
There is profound skepticism toward this idea. Like, it’s a non-starter. One expert says it would be easier to operate data centers at the bottom of the ocean. There are a lot of technical barriers to running data centers in space, such as radiation that could fry chips unless they’re hardened and latency lags that could make the centers unreliable.
It seems far more plausible that technological advances and possibly game-changing breakthroughs will improve the efficiency of terrestrial data centers, leaving space to the satellites. The big companies powering the AI boom have thought about all this and developed elaborate long-term strategies that are unfolding gradually. Don’t underestimate them.