Silicon Valley called — the 1990s are back

Rana Foroohar rana.foroohar@ft.com · 27 Oct 2025

San Francisco has been economically detached from the rest of America since the internet boom of the mid1990s. As everything from housing rents to per capita growth rates far higher than the US national average show, the Bay Area exists in its own orbit.

I couldn’t help but think of this when I flew into San Francisco a couple of weeks ago. Artificial intelligence, rather than consumer web pages, drives the exuberance today. But the vibe is the same as it was 30 years ago — there is one story, and everyone is buying it.

Every single billboard I passed from the airport to the city had something to do with AI. Giant white blocks of brandnew housing and office space lined the highway. In the windows, I saw young men (Silicon Valley employees are still mostly men, just like back then) typing away at computers while sports or video games played on their oversized flatscreen televisions.

So far, so 1990s. Which leads me to wonder what’s different this time around? It’s a topic that has been much discussed in recent months as it has become clear how much of the US growth picture depends on AI.

While some of the answers are easy — most people agree that the long-term impact of AI will be far broader and deeper than the rise of the consumer internet, even if we see a short-term correction — I would point to three comparisons between the 1990s and today that warrant more investor consideration.

First, AI is highly capital intensive. The dotcom bubble wasn’t. Back then, anyone could throw up a homepage with almost no investment. Today, AI investment represents all the growth in corporate America’s capital expenditure.

Contrarians might point out that there were more than 80mn miles of fibre optic cable laid from the mid1990s until the end of the dotcom boom, and much of that investment took years to pay off. US telecom companies spent $444bn in capital expenditure between 1996 and 2001.

Still, compare this with the $342bn that will be spent this year alone in the US by the top investors in AI data centres and computing infrastructure, including Microsoft, Alphabet, Amazon and Meta. At the current rate of power consumption needed to fuel AI development, estimated investment will stretch to nearly $7tn by 2030.

This brings us to another key point on the capital investment front, which is that energy is a hugely significant constraint on the development of AI today. That was not true back in the 1990s.

If the current energy economics of AI hold, then the investment story becomes even more costly and complex, since it is dependent on the trajectory of grid upgrades, more nuclear power coming online, how quickly renewables can replace fossil fuel and how high oil prices will go amid these changes.

Another difference between the two eras is that AI investment today is increasingly funded not by debt, but by equity. At first glance, that seems like a very good thing. The half a trillion dollars in telecom investment during the dotcom bubble was funded mostly by debt, whereas the big AI players today are huge, profitable companies.

For that reason, we probably won’t see a repeat of failures like WorldCom or Global Crossing, since the Magnificent Seven (or perhaps four, when it comes to AI) are not about to go broke. The richest companies in the history of global capitalism have cash to burn.

But this leads to what may be the most important comparison between the 1990s and today.

Many market participants argue that we shouldn’t worry so much about the fact that AI hype seems to be holding up the entire US equity market, since the earnings of the Big Tech companies driving the boom are so huge.

Yet when you look beneath the hood, that’s not the full story. Apollo’s chief economist Torsten Slok, who is known for his ability to land on the metrics that really matter, put out a remarkable chart a few days ago. It showed that since Donald Trump’s tariff “liberation day” in April, companies in the Russell 2000 with negative earnings are outperforming those with positive earnings.

This small-cap index includes many companies in AI-investment heavy areas such as software, biotechnology and healthcare. But it also includes plenty of others in sectors such as financials, energy, materials and communications, which are similarly betting on an AI future. It’s an uncomfortable rhyming of history with the dotcom boom, when negative-earnings companies also over- took those with positive earnings for a prolonged period.

What should we take from all this? A few things.

First, although “AI will be a revolution in so many ways, there is an echo effect from the 1990s that we should pay close attention to”, says Slok.

What is more, the economic impact of even a short-term collapse in the AI bubble may be bigger than it was in the dotcom era, given that more Americans now hold more stock than ever before.

Second, the dynamics underlying this bubble are far more complex than they were two decades ago. Will produc- tivity gains offset AI-related job losses? Will energy remain a chokepoint? Will a Chinese innovator disrupt the market? Or will US growth continue to outpace the world, as it did in the 1990s?

Prospects for more than just Silicon Valley hinge on the answers.