Speaker A:
About a week ago, we made a video on TLDR Global explaining the so-called circular financing of AI companies. Before that, we made a video on the hidden debt being accrued by hyperscalers.
It's true that risk has built up in the financial system that underpins AI. But the lingering question about what could trigger a collapse in its value lies outside of the realm of financial instruments. It's a pitched battle in the real world, influenced by the attitudes of local communities, the leaders they elect, and the decisions they make.
In this video, we're going to look at public opinion towards AI, its political impact, and finally, at the way all this is impacting tech markets and companies. All our evidence will suggest that politics, instead of pure economics, could burst the AI bubble.
The future war between humans and AI might have already started, and it's causing big tech companies to rethink everything.
In our new magazine, Too Long, we explain how blocking new data centers might finally burst the AI bubble. That's just one of the articles in our longest issue ever. Subscribe to get your copy by clicking the link in the description.
Before we start, a quick note on why the US is at the centre of this debate right now.
The US blows every other country out of the water in terms of how many data centers it has and how many it's building. It's also the centre of AI investment, again by a huge margin.
This is where the so-called AI bubble is growing, and where it could pop.
So let's start on the ground and ask how the people at the centre of the AI buildout feel about data centers.
Our source here is fresh polling data from Heatmap, which has drawn a lot of attention from politicians and the media.
What you can see here is that since August last year, public opinion on data centers has become dramatically worse.
This time last year, the majority of people supported or were ambivalent towards data centers being built in their local area. Today, 75% of Americans polled oppose building them, with most of that group strongly opposed to data centers being built near them.
So why the sudden backlash?
Well, some of it is because of what we might describe as their polluting effects. Data centers take up a lot of space, use up a fair bit of water, and use a lot of electricity, which pushes up power prices.
That on its own is enough to make them unpopular.
But on top of that, there's a widespread sense that AI is just unnecessary or even harmful.
AI's PR problem hasn't been helped by comments from big tech CEOs.
For example, back in 2023, Sam Altman was very clear about how AI might affect employment. He said in The Atlantic that jobs are "definitely going to go away, full stop."
In the same year, Elon Musk said there will come a point where no job is needed.
While business leaders in the Bay Area might be thrilled by the idea of replacing their workforce with robots, the average American doesn't seem to think the same way.
This makes data centers unlike other bits of infrastructure like, say, car factories.
Like data centers, car factories are polluting in lots of ways, but residents are generally more likely to tolerate them because most people believe you need cars.
Furthermore, unlike traditional factories and other infrastructure projects, data centers generally don't create too many local jobs.
The construction work is usually so specialist that you need to bring in outsiders, and most jobs that are created are temporary. After all, once a data center is built, you don't really need anyone to run it anyway.
Heatmap's polling also shows that the Democrats and Republicans share this antipathy towards the AI buildout.
So now we're in a situation which nobody saw coming, where the midterms have become a battle in which candidates across the political spectrum compete to distance themselves from AI and data centers.
It doesn't matter how much Trump likes AI or how much money Wall Street has bet on it, because individual midterm candidates are fighting tooth and nail for their careers.
And especially in states with a lot of data centers, they're scared of losing ground to popular anti-AI opposition.
There's evidence for this.
Look at this private memo from the National Republican Senatorial Committee, which they sent to AI companies to warn them about the situation on the ground.
There are a few lines which really stand out.
The Republicans worried about their candidate, John Husted, and his chances in the Ohio senatorial election, with his opponent, Sherrod Brown, making data centers the centre of his campaign and polling well for it.
They say AI opposition has become a sleeper issue for the entire election cycle.
Their warning is dire: if Husted loses because of data centers, the industry will be blamed, and other candidates will copy Brown's strategy to win their own elections.
To some extent, that's already happened.
At a glance, you can see that data centers are a sticking point in the senatorial or gubernatorial elections for over a dozen states, including Iowa, Georgia, Michigan, Texas, and Maine.
Elections aside, 25 of America's 50 states are weighing up anti-data-center legislation, and more than 211 localities have already taken action.
So what impact could this have on the markets?
Well, any delay to a data center being built is hugely costly.
A report from the Carnegie Endowment found that a construction delay of just one year reduces AI data center value more than electricity costs doubling.
Multiply that out by the number of data centers being built, and the dollar exposure shoots up to figures in the hundreds of billions.
They might be able to find other places to build their data centers instead, but again, this would be costly and time-consuming.
Lots of these companies have also issued debt on the assumption that their data centers would be built in a few years.
If this doesn't happen, they'll probably have to refinance at longer maturities.
This would be bad because borrowing costs have already gone up, presumably in part because, as the political tide turns, creditors are becoming worried about regulatory risk.
Back in July, for instance, Bloomberg reported that 80% of data-center securities sold since early last year are now quoted at a wider credit spread.
With slightly less jargon, this means that, measured against the baseline of U.S. Treasury bonds, the cost to companies of borrowing money for data centers has increased.
On top of this, we can look at Anthropic, which is poised to go public very soon.
Right now, they're having meetings with bankers and investors in San Francisco, trying to drum up hype for their stock for when it's launched.
Sources familiar with these meetings told CNBC that investors are asking about the financing impact of a slowdown in data center building.
In other words, investors are trying to factor these new risks into the price of major AI companies, and it's something we'll be able to see more clearly when they go public.
As a final thing, when people talk about the AI bubble, they're usually suggesting that, as a technology, AI isn't all it's cracked up to be.
Hence, the argument goes, all the trillions being poured into it won't ultimately be made back.
This data center backlash creates a sort of second bubble-like risk.
Even if AI ultimately turns out to be the world-altering technology its proponents claim, if politics gets in the way of the data center buildout, there's enough time-sensitive debt floating around that you could still see a bubble pop.
I've already told you about one of the articles in the next issue of Too Long, but there's so much more inside you're going to want to discover.
If you subscribe using code SUMMER26, you'll not only get 20% off every issue as a subscriber, you'll also get an additional £3 off your first issue, bringing the cost as low as £4.99 for a digital edition.
And yes, if shipping or increasingly tariffs are making deliveries to wherever you live a little too expensive, you can get a digital version sent straight to your inbox.
Plus, we actually earn more from digital editions, as we actually lose money on shipping, so it's also a better way to support the channel.
This is a good first issue to jump on board with too.
It's our longest ever — 84 pages — and it contains stories from all across the world, including Burnham here in Britain, the wars in Ukraine and Iran, China and the US's relationship, a bit from the Vatican, and, of course, the COVID story on the plan to rebuild Europe.
It also contains articles from our superb new writers, Natasha Isaac and Ebba.
So make sure to use code SUMMER26 at checkout and become a Too Long subscriber today.
Comments
Glynnis Campbell
This is a test comment!