Friday charts: Underestimating tomorrow

Thirty years ago this month, Mary Meeker published her first annual Internet Report. 

February 1996 was still early enough that Meeker, even with her an audience of sophisticated investors, had to start by telling them what the internet was:  

Many of the predictions she proceeded to make now seem laughably obvious: 

It’s hard to imagine that, six months after the Netscape IPO, those qualified as predictions worthy of a Morgan Stanley research note.

But investors still needed convincing that this “internet” thing would be relevant to them; so she included some anecdotal evidence that it was really catching on:  

The tech-stock boom was already underway by then: Meeker noted that the five stocks in her Internet Portfolio were up 257% in 1995.

But Main Street wasn’t feeling it yet. “It is tough out there,” Meeker empathized. “Layoffs, stagnant real wages, and longer work days are only part of the malaise that is gripping the American worker/consumer.”

Things were so tough that, to try out the internet, many had to economize: “When they find something new they like (i.e., computers) they must give up something in return (i.e., cars),” Meeker said. “Consumers are buying PCs instead of cars,” her economist colleague Steven Roach confirmed.

Now, when we remember the ‘90s, we’re really only remembering the second half of it, when tech stocks were booming and unemployment fell to a 30-year low of 3.9%.

Meeker became a symbol of those late-90s boom times — the “Queen of the Net,” as she came to be known.

Put on a pedestal like that, she inevitably came to be associated with the bust, as well. “Celebrity Internet analysts such as Morgan Stanley’s Mary Meeker,” PBS opined in 2002, “have come to symbolize the intellectual hollowness of the Internet bubble’s headiest days.”

This proved to be unfair, because headier days were coming — headier even than Meeker expected. “While many of her boldest projections seemed wildly optimistic at the time,” a recent study concludes, “the vast majority proved not only directionally correct but often conservative in hindsight.”

That shouldn’t have come as a surprise. Way back in 1951, Arthur C. Clarke told us to expect new technologies to exceed even our most daring prophecies.

It’s happening again. 

In 2022, a survey found that AI experts expected the “full automation of labor” to occur in the year 2164.

A year later, after the introduction of ChatGPT, the same survey expected full automation to occur in 2116 — 48 years earlier. 

Who knows what they might expect by now. But, whatever our predictions are for AI, they will probably prove laughably conservative.

That will sound like bad news to the many investors who got spooked by the “AI scare” — this week, the stock market worried that AI will be too good, putting software companies out of business and people out of work.

But Meeker advises optimism.

In a report on AI — still styled just like her Internet reports of 30 years ago — Meeker again took the long view: “Time and time again, the case for optimism is one of the best bets one can make.”

I predict that prediction will age well, too.

Let’s check the charts.

How it started:

Meeker used this graphic of “Internet hosts” in her first report to illustrate the rapid growth of the internet — from a few hundred in 1986 to nearly seven million in 1996. She defined an Internet host as any device connected to the internet, so that number is now in the billions.

How it’s going:

In 1995, Meeker optimistically predicted internet users would be 7% of the world population by 2010. The actual number was 29%. Today, it’s 60% (which must be nearly 100% of adults).

It’s different this time:

The dotcom bubble was fueled by debt, while the AI bubble — if that’s what it is — is being fueled by cash flow (so far).

Under-optimistic?

Goldman Sachs estimates that investors are expecting S&P 500 earnings-per-share to grow at 10% over the long term — higher than average but well below the dotcom bubble.

More balanced:

Global tech stocks trade at just a 30% premium to the stock market average — modest compared to the 130% premium it hit in 2000.

AI ramping up:

Ramp reports that more than one-third of the businesses using its expense-management software are subscribed to ChatGPT, and nearly one-fifth (and rising) are subscribed to Claude.

Anything but tech:

It’s only February, but the equal-weight S&P 500 is outperforming the market-cap-weighted S&P 500 by even more than it did in the dotcom bust.

Not very bubbly:

Correlation between Hyperscaler stocks is low, which is not the kind of behavior you’d expect in a proper bubble. It’s when they’re all moving in the same direction that you most have to worry.

Things happen fast:

One measure of AI models is how long they can work on a coding task. The models were improving exponentially, doubling the time they can run every seven months (the orange line). Now it’s doubling every four months (the red line). AI Digest calls this the new Moore’s Law for agents.

The internet in 1995:

37 pages of Mary Meeker’s 323-page Internet Report were dedicated to suggesting websites for investors to try out. Few have survived, and most look pretty comical now. I’m sure today’s AI will look even more so to the people of 2056.

“The magic of watching AI do your work for you,” Meeker says now, “feels like the early days of email and web search.” 

The difference this time? “The better / faster / cheaper impacts of AI seem just as magical,” she concludes, “but even quicker.”

Have a great weekend, prophetic readers.

— Byron Gilliam

_{Brought to you by:}

AI hits a power bottleneck. Bitcoin miners have the solution: plug-in ready energy grids.

Miners are pivoting to High-Performance Computing, leasing grids to AI hyperscalers. VanEck Onchain Economy ETF (NODE) captures this convergence, holding the infrastructure bridging digital assets and the AI grid.

Crypto’s premier institutional conference is back this March 24–26 in NYC.

Don’t miss SEC Chairman Paul S. Atkins’ keynote on Day 1.