How to Navigate a Tech World Dominated by AI – Uncharted Territories

This post was originally published on this site.

AI is disrupting the world of tech. How can you navigate it:

As a stock market investor?
As a VC?
As a tech employee?

These are the three topics what we’ll cover today.

None of the content in this article is financial advice. I am not a financial advisor. I’m just musing and speculating based on my personal experience. Please don’t make investment decisions based on this, and please consult your registered financial advisor, since a government stamp of approval is certainly the strongest indicator of quality financial advice that could possibly exist.

This is what you and I do when we put our money in a 401k, or a robo-advisor, or straight into the stock market. And to speculate on what will happen here, it might help to understand what happened in comparable situations in the past.

In the 19th century, there was a frenzy of railroad building in the US. Hundreds of companies wanted in, and most lost money, even as the country’s citizens experienced a golden age of travel. Why?

Building railroads required massive upfront investments in infrastructure like tracks, bridges, tunnels, and stations, which strained many companies financially. A lot of capital was sunk before any returns could be realized.

After the initial boom, competition became fierce. There were often too many rail lines built, leading to overcapacity in some regions. Customers could use competition between providers to pressure prices down. This forced companies to cut prices to attract customers, reducing profitability.

But in places where companies had monopolies because they had been the only ones to lay down railroads, they could charge customers much more. As railroads became more vital to national infrastructure, governments began imposing regulations (such as rate ceilings) to keep prices low. The problem is that companies had lines with monopolies and others with competition. With regulation, the monopolistic lines stopped funding the loss-making ones, further squeezing profits of the parent companies.

Since railroads were capital intensive, but they were made of physical infrastructure that was easy to recover, interest rates on loans were lower than in comparable industries. Many railroads used that to over-leverage, and a lot of this investment was speculative, which was catastrophic in speculative bubbles such as the Panic of 1873, which led many railroads to go bankrupt.

A bank run on the Fourth National Bank in the Panic of 1873

The large network of railroads coupled with low fares meant customers—both passengers and freight shippers—enjoyed a lot of the surplus. The result is that this industry created lots of value for customers, but little of it could be captured by investors.

To be clear, this is capitalism nearly at its best: Lots of value generated for everybody, little resulting inequality. It’s good—unless you’re an investor.

Something similar happened with airlines.

Like railroads, airlines require huge upfront capital investments—aircraft, infrastructure (airports, terminals), and maintenance. Additionally, operating costs (fuel, labor, maintenance) are high, leaving little room for profit margins.

At the same time, since the industry is customer-facing and passengers are extremely price-sensitive, airlines have often engaged in price wars, especially after deregulation in the U.S. in the late 1970s. Low-cost carriers intensified competition by offering cheaper fares, forcing established airlines to lower prices, reducing profitability.

In this type of situation, the smallest hiccup can cause serious damage. Unfortunately, the industry is highly cyclical and vulnerable to external shocks such as fuel price spikes, economic downturns, and events like 9/11 or the COVID-19 pandemic. This has resulted in many airlines going bankrupt, being forced into mergers, or requiring government bailouts. Examples include PanAm, TWA, Eastern Air Lines, Swissair, Sabena, Continental, Alitalia…

Of course, at the same time, customers have reaped huge benefits from this competition. Airfares (especially adjusted for inflation) have generally trended downward over the decades, making air travel accessible to more people. The overall quality of service, frequency of flights, and access to international destinations have vastly improved. The result is that customers have benefited from globalization, economic integration, and personal mobility at unprecedented scales, but airlines often struggle to capture a significant portion of this surplus as profit.

This is not what you wanted. You wanted to be crammed in a sardine box so that you could shave a few dollars off your ticket. Admit it.

Will the same happen with AI foundation models?

Foundation models are the software that power OpenAI’s ChatGPT, Anthropic’s Claude, Meta’s Llama, Google’s Gemini, and the like. It’s very expensive to make them. Today, it’s in the order of hundreds of millions of dollars. In the not-too-distant future, it will likely reach billions, and within a decade, it might reach a trillion.

This is certainly a big barrier to entry, and many hope that this will create a deep moat that will make the companies that are able to train these models very valuable. But will they be, or will most of the value end up in the hands of consumers, like with railroads and airlines? I don’t think they’ll make that much money.

First, these models are extremely valuable, but the most valuable companies in the world know this, and they’re all competing to make them. Microsoft, Amazon, Alphabet, Meta… They’re all in. They know if they don’t compete, they might be left in the dust.

Most notably, Meta has PTSD from the mobile years, when Apple and Google treated it as their bitch: They told it what it could and couldn’t do on the App Store and the Google Play store. This is the main reason that Zuckerberg invested so much in virtual reality: He wanted to own the next platform. Now, with AI, he’s taken the lesson from Google, which made Android open source to control the platform (and defend against Apple). Meta is doing the same, and is releasing its Llama models for free in open source. And they’re not bad! They compete head to head with the best models from OpenAI and Anthropic.

These models also turn out to be quite comparable in quality. It’s not like OpenAI’s is 10x better than Anthropic’s.

Here, I ordered the models by index, which conveys their quality. As you can see, there are many models that are comparable in quality. There are none that blows the others out of the water. Source.

So to summarize, in this industry:

There are many players
They need lots of money, but it is there for this
Some very deep-pocketed players would rather die than stop competing
Quality is not a differentiator
At least one of these big players is willing to give the stuff away for free

So, personally, I think foundation models will go the route of railroads and airlines: I commend the investors who are burning cash for the sake of humanity.

This means, though, that these companies won’t have outsized returns, and your investments in the Vanguard ETFs tracking the NASDAQ and S&P 500 are not any safer.

In the 1848 gold rush to California, most gold diggers didn’t make much money, but the shovelmakers made a fortune. NVIDIA is today’s shovelmaker.

When we say “it costs billions to train AI models”, what we mean is “all these AI companies are spending billions mostly in a bidding war to get chips from NVIDIA, which they then use to train the models.”

That’s why NVIDIA is one of the most valuable companies in the world right now:

Note that many of the top 10 are training their own models, by spending money on NVIDIA. Meanwhile, NVIDIA spends money on TSMC at the bottom, which is the company that actually manufactures the chips. Source.

NVIDIA has a big advantage over the other companies, but that’s because they’ve been designing these chips for a very long time. Now that they’re so valuable, companies like Meta and Google are racing to design their own. And since TSMC doesn’t design its own chips, it’s happy to manufacture anybody’s design. As a result, my guess is that NVIDIA’s lead will erode over time.

Right now NVIDIA is so valuable because there’s a frenzy of demand for its product, and they have a quasi-monopoly on its supply. They’re the Saudi Aramco of the 21st century, but instead of black gold, they have silicon gold.

So the question becomes: What will grow faster, the supply of its competitors, or the demand for more model training? To give you a sense, its revenue more than doubled year over year in August, but that was not enough to satisfy investors, and the stock has lost 17% of its value since the news was made public. In other words, for investors, only doubling revenue was a sign of future weakness.

My guess here is that competitors might not immediately be able to do what NVIDIA does, but they’re going to close the competitive gap over time, which means NVIDIA won’t be one of the most valuable companies in the world for decades to come, and we can’t count on its value to keep growing. Retail investors won’t be able to trust their retirement on this shovelmaker.

Where to go, then?

Here’s the problem: The stock market has grown handily over the last couple of decades, but most of this growth was on the back of tech companies. When you look at the stock market without tech, an ugly picture emerges:

If this wasn’t enough, tech valuations are getting closer to where they were before the 2022 crash. The only hope here is that interest rates are still high and could go down, which would push stock prices back up. But outside of that, I don’t see great news for retail investors. If we summarize this article and the last one:

Fewer tech companies will go public.
Those that do won’t necessarily be the best.
The good ones that still raise money from the stock market will likely raise less than in the past.
More competition is appearing due to AI that will likely undermine the value of public companies. A sizable chunk of this competition will come from companies that will never be public (or that will take long enough that most of their returns will have been earned while private).
Existing top tech stocks are fighting the foundation model war, but it’s unclear if any of them will capture a lot of that value.
NVIDIA is unlikely to keep growing forever.
Outside of tech, the stock market is not that productive. We can’t expect high returns from there.
Valuations are already quite high.

So what should you do as a retail investor? As a rule of thumb, it’s probably good to remain diversified. But where I thought annual returns of 5-10% on average were something to expect in the long term, I’m less confident now—especially in the tech industry. I used to have the lion’s share of my money in tech stocks, but now I’m rethinking that. Maybe I should diversify more now.

Outside of tech, public companies that operate in highly regulated markets or very physical ones are probably safe investments, but they’re not super likely to provide shining returns. Who wants to bet on Boeing and LVMH for their retirement?

OK, let’s now look at VCs and tech workers.