Michael Burry – the man who famously spotted the US subprime mortgage crisis years before most others – thinks there's an AI bubble.
This week it was revealed that he had taken a $1.1 billion bet against chip-maker Nvidia and Peter Thiel’s Palantir – where he will profit if their stock prices fall.
But it’s not just professional market sceptics like Burry that think things may be overheated in tech.
Jamie Dimon, the head of JP Morgan, recently warned that US stocks – in particular tech stocks - are over-heated. Morgan Stanley chief executive Ted Pick and Goldman Sachs head David Soloman have also warned of an impending correction.
Even big tech cheerleaders like Amazon founder Jeff Bezos have said there is an AI bubble. Meanwhile Sam Altman, the head of AI revolution icon OpenAI, thinks valuations and spending has gotten carried away.
Why do they think this is a bubble?
You can get some clue of why if you look at the quarterly results some of the world’s big tech companies published last week.
Microsoft, for example, said it spent nearly $35 billion on capital expenditure in the quarter. Capital expenditure wouldn’t necessarily all be AI-related but in this case the vast bulk of it is.
To put that figure into context, Ireland’s Budget 2026 was worth around €9.4 billion - or roughly $10.8 billion. So Microsoft spent more than three times Ireland’s annual budget in a single quarter.
Meanwhile Facebook owner Meta said it would spend as much as $72 billion on capital expenditure this year. Google owner Alphabet said it would spend up to $93 billion. Amazon said its full year capital budget would be $125 billion this year – and even higher next year.
And that’s just four of the big tech companies - it wouldn’t include AI spending by other large firms like Apple, Tesla, X, Oracle or Palantir. Nor would it include the tens hundreds of millions - or even billions - being spent by other big-but-not-massive companies.
Or the money being invested by venture capitalists and private equity.
And most of that money is ultimately going towards building data centres and buying high end computer chips - but it’s also being spent at employee level too.
Back in June Sam Altman claimed that Meta was offering his top engineers pay packets valued at $300m if they moved over to Team Zuckerberg. That included a $100m signing-on bonus.
The likes of Meta have also spent billions to buy stakes in, or acquire outright, AI startups that have promising-looking ideas.
The scale of all of this is so significant that Goldman Sachs estimated that AI investment was a bigger contributor to growth in the US economy in the first half of the year than consumer spending.
In fact, if you took out the money being spent on AI, the US economy was stagnant in the first half of the year.
And tech companies’ valuations have risen to match – largely based on the promise and expectation that all this spending now will result in them making even more money in the not-too-distant future.
Are there any signs of that happening?
No - An MIT study in August found that, when you strip away all the hype, 95% of firms were getting zero return on their AI investments.
It’s not entirely true to say that no-one is making money from AI… some are. Or, more specifically, one company is.
So far the only clear winner from the AI race is Nvidia, which designs the computer chips that are powering all of these data centres that big tech firms are rushing to build.
It’s gone from being a relatively niche player in the market to being the most valuable listed company in the world.
As recently as its 2020 results, Nvidia had profits of $2.8 billion – which isn’t bad by any measure. But now, in the single quarter between May and July, it had a profit of $26.4 billion.
It is now comfortably on track to make more than $100 billion in profit this year – which would be about $36 times the profit it made just a few years ago.
For most others, though, it’s a different story.
Despite their huge investments to date the likes of Meta, Alphabet and Microsoft have yet to really show the benefit AI has brought to their bottom line.
Some are pointing to it being a cost-saving measure – Amazon’s recent jobs cuts, for example, were framed as them leaning on AI to be able to reduce bureaucracy. But if that’s the rationale it will take many more cuts to justify the hundreds of billions of investment.
And it’s not just the big tech companies that are struggling to make it viable – the dedicated AI creators are not seeing a return yet either.
Anthropic made a $5.3 billion loss last year. Elon Musk’s xAI is predicted to record a $13 billion loss this year.
Even the poster-child of AI – OpenAI – made a $5 billion loss last year.
That means that every time you ask ChatGPT to compose an email for you, or as it to devise the best scrambled eggs recipe, it’s costing them money.
Despite that, there are suggestions it could be valued at $1 trillion when it floats on the New York Stock Exchange in the not-too-distant future.
So if it’s not making money, what’s the justification for that valuation?
OpenAI and its backers hope it is following the same kind of playbook we’ve seen from so many tech start-ups over the years.
This is where it has a great idea, great technology and great people – and once it builds a big enough customer base and gains efficiencies of scale, the profits will start to flow.
And, in fairness, its revenue looks to have more than doubled year on year – and it recently started to unveil a range of products and services it plans to provide – including a web-browser, which will open the door to it service ads too.
In the meantime, though, they’ve also committed to invest around $1.4 trillion in other companies over the next few years – which will include buying computer chips and data centre capacity to help handle the queries from all of those extra customers.
But within those deals it’s making is also an interesting dynamic that has added to concerns about an AI bubble – and that’s how a lot of the investment that’s taking place seem to be almost circular.
For example, one big part of OpenAI’s recent investments was a $300 billion deal with Oracle to tap into its data centre capacity.
In turn, Oracle has committed to spend tens of billions of dollars to buy Nvidia chips to power its data centres.
Meanwhile, Nvidia has invested $100 billion of its money into OpenAI.
Which essentially means that the three companies are passing billions of dollars around in a circle.
So if this is a bubble, what happens when it bursts?
Well the first thing there is the 'when’ of the when. Just because there’s a growing consensus that there is a bubble, doesn’t mean that it’s ready to pop.
Going back to Michael Burry – he made his first bet against the subprime mortgage market in mid-2005; it took pretty much two years for that bet to pay off.
And as readers (or viewers) of The Big Short may remember, he received a lot of flak during those two years from investors who thought he was costing them a fortune.
It should also be said that, while he was right that time, he can be wrong too. He took bets out against big tech stocks in 2023 too, and the tech-heavy Nasdaq has risen by nearly 80% since then.
But assuming it is a matter of ‘when’ and not ‘if’, there are a whole range of predictions around what a burst AI bubble might mean.
Meta CEO Mark Zuckerberg has kind of downplayed his company’s exposure - saying that spending all of this money on data centres now was essentially just front-loading what they’ll need in the years ahead.
So, he argues, worst case scenario it’ll just mean they’ll be futureproofed against a natural growth of demand for its services in the years ahead.
Amazon’s Jeff Bezos suggested that while this is a bubble, it wouldn’t be a ‘bad’ bubble like the financial crisis. Instead he said it would be a ‘good’ bubble, comparing it to the Dot Com crash.
His argument is that while a lot of companies collapsed in the Dot Com crash, a handful that had good ideas and viable models – including Amazon and Google – survived and thrived. He thinks the same will happen here – the tide will go out and expose those that can’t back up all of their big talk, but the useful platforms will stick it out and usher in big changes for the world.
Of course that comparison does skirt over the negative impact this kind of ‘good’ bubble can have. All of those bankruptcies and valuation write-downs means lost investments and a shock to the stock market, which perhaps impacts people’s pensions, and likely leads to job cuts too.
So it wouldn’t just be a case of adjusting a balance sheet or two and moving on.
But is there the potential for contagion?
Again, it depends first and foremost on how much more bubbly the bubble gets before it bursts.
But no more than AI investment is feeding growth in the wider US and world economy right now, a dramatic collapse in investment and valuations in the sector will have a knock-on effect on national and global GDP too.
There’s also the very practical impact of jobs cuts – direct and indirect.
If it turns out that the AI industry has gotten too big and expensive, some of those being hired now will likely be let go. Meanwhile companies that see the value of their AI investments getting written down may need to make cuts elsewhere to cover their losses.
It could also have a knock-on impact on areas like construction and consumer spending.
And one concern around a potential AI crash relates to that circularity of investment that’s being seen at the moment.
A big part of the boom in Nvidia’s revenue is based on the scramble to make use of OpenAI’s tools. Oracle’s stock price also reflects the fact that it’s been promised $300 billion of business from OpenAI in the coming years. As a major shareholder in OpenAI, Microsoft’s valuation is also boosted by the belief that Sam Altman’s firm will become massively profitable in the near future. Meanwhile Amazon has done its own multi-billion dollar deal with OpenAI to provide it with data centre capacity to power its AI platforms.
That means that all of those big tech companies have a lot to gain from OpenAI’s success – and a lot to lose if it turns out that the technology isn’t nearly as useful as many hope.
Its importance to these major firms has led to suggestions by some that the company is already too big to fail. Some have even started to call for it to be broken up to reduce the systemic risk it poses.
That’s despite the fact that it’s currently still privately held, and still losing billions of dollars a year.
