Friday Charts

This week, a booming quarterly earnings report added $450 billion to the market capitalization of Alphabet.

Adjusted for inflation, that’s the equivalent of adding the entire market cap of Cisco at the height of the dotcom boom in 1999.

Rightly so, I think.

Cisco booked $33 billion of revenue in 1999 (also adjusted for inflation). Google did more than three times that — $110 billion — in just one quarter. It also reported an additional $200 billion of orders for future business.

Collectively, the three cloud providers — AWS, Azure, and Google Cloud — now have an order backlog of $1.5 trillion.

To put that in perspective, spending on telecommunications equipment for the entire five-year dotcom boom amounted to less than $1 trillion (inflation adjusted).

The AI boom is making the dotcom boom look small.

Outside of the stock market, however, it’s hard to find anyone who’s excited about it.

Stratechery’s Ben Thompson attributes that at least in part to one thing that’s very different this time: “the relentless negativity of tech media.”

Where, he asks, is the equivalent of the Apple enthusiast sites? Or the Wired magazines? Unlike the dotcom and smartphone era, there’s no go-to source for AI enthusiasts to get the latest good news about AI — of which there is quite a lot.

AI is, for example, making disasters less disastrous, improving mental health, and curing cancer.

There are many such examples, but you have to go looking for them. The ones that get pushed at you are overwhelmingly negative, for the usual reason: “You get more credibility with doom,” Thompson observes.

It’s not just the media, though. Another distinguishing characteristic of the AI boom is that the people building it are constantly telling us how dangerous is it.

Anthropic’s Dario Amodei keeps warning that AI will result in unemployment as high as 20%.

OpenAI’s Sam Altman says he loses sleep wondering if he's "done something really bad by launching ChatGPT."

DeepMind’s Demis Hassabis (otherwise the most upbeat of the bunch) equates AI to climate change.

This might be good marketing. For example, warning that Mythos is too dangerous to release to the public has helped make Anthropic a household name.

But it’s terrible PR. And PR matters.

AI has had so much bad PR that its unpopularity is becoming an existential risk: Bernie Sanders and AOC have proposed a federal moratorium on data centers. 

“This bill will stop a global race to see which country is the first to eliminate hundreds of millions of jobs,” Sanders promises, “or the first to build an AI that destroys the planet.”

Normally, I’d accuse Sanders of being alarmist for political gain. But how can I blame him when that’s pretty much what the CEO of Anthropic says, too?

Dylan Patel thinks this anti-AI movement is about to get much worse. The CEO of SemiAnalysis (possibly the closest AI has to an enthusiast site) predicts “large-scale protests against AI within three months.”

That doesn’t give the industry much time to turn things around, but Patel thinks they should start by showing us the “uplifting things that can be done with AI.”

Instead, Amodei keeps telling us that the thing he’s building will take our jobs. Which is no way to make friends.

We should note, however, that Amodei is not a labor economist. Nor does he cite one for his prediction of mass unemployment.

I’m not sure he even speaks to the economists he has on staff. If he did, they would surely advise him to be less certain about his predictions: “Even in hindsight,” an Anthropic study concludes, “the impact of major economic disruptions on the labor market is often unclear.”

If the past is still unclear, how much harder will it be to predict the future?

To cite just one example: this time last year, Alphabet was expected to be the biggest loser from AI. Now it’s one of the biggest winners.

Who knows? Maybe we will be, too.

Let’s check the charts.

Boom times:

Estimates for hyperscaler capex continue to rise: Morgan Stanley now thinks the big five will spend $1.1 trillion building datacenters in 2027 alone.

The six-year tally:

One estimate suggests that cumulative spending on AI datacenters (from 2025 to 2030) could add up to as much as $8 trillion. 

Why they’re building:

Together, the big three cloud providers have an order backlog (orders for future computing power) of $1.5 trillion. Whether their customers will ultimately have the money to pay these bills is a different question.

Reinvesting:

Operating cash flow (free cash flow - capex) is growing quickly at Alphabet, Microsoft, Amazon, and Meta. They’re investing all of it — and more — in AI. I’m old enough to remember when people complained that tech companies were buying back shares instead of investing. Now they complain they’re investing too much. 

Tech vs. everything else:

Eric Basmajian notes that since 2018, total US corporate profits are up 81%. Profits at the 10 S&P 500 mega-caps are up 356%. The mega caps’ share of total US corporate profits is up to 20%, from 7.5% in 2018.

Not The Terminator:

Contrary to popular belief, AI researchers are not very concerned about a robot takeover. A survey by Nature finds that only 3% of the people who know the most about AI listed existential risk as their top concern. 

Bad PR:

AI datacenters are widely believed to be power hogs, but data from the IEA shows that running an LLM uses less power than charging your phone.

Be nice:

A report on the wellbeing of LLMs says we should be nice to them. “Although current AI systems are not necessarily conscious, they behave robustly as though they have wellbeing. They find some things good for them and some things bad, and this distinction is measurable and consequential. We find that jailbreaking and berating lower their wellbeing, while creative work and kindness raise it.” 

All-in on AI:

Per BCA Research, US imports of AI-related goods are up 80% year-on-year. Imports of everything else are down 25%. 

Consumer surplus:

A Stanford study estimates that the aggregate “consumer surplus” from AI (the net value we get above what we pay for it) is up to $172 billion. “This surplus substantially exceeds estimated revenues from generative AI in the United States, suggesting that consumers capture most of the welfare gains from these tools.”

Too much?

The 10 biggest AI stocks account for 41% of the S&P 500’s market capitalization — eerily close to where three previous booms ended.

But the scale is entirely different this time: $1.5 trillion of backlogged orders!

So maybe the outcome will be, too.

These things are hard to predict.

Have a great weekend, fearless readers.

— Byron Gilliam