AI Anxiety Reaches Fever Pitch. Why It’s Still Not Time to Worry.

SoftBank CEO Masayoshi Son needs to fund his firm’s $22.5 billion investment in OpenAI. He’s using Nvidia stock to get there. (Kiyoshi Ota/BloombergBloomberg)

The market action around the AI trade—plays on the global AI investment boom—got even choppier this past week. Major reactions to relatively minor events show that shareholders are getting antsy as AI investment grows with little measurable return behind it.

The talk of an investment bubble is getting louder, and there seem to be a lot of people with one foot out the door, ready to jump at a moment’s notice.

The worries began with a Monday X post by Michael Burry of The Big Short fame, who shorted mortgage-backed securities leading into the financial crisis. Two weeks ago we learned that Burry had placed big short bets on Nvidia and Palantir Technologies during the third quarter.

In his Monday social media post, Burry claimed that Big Tech companies that are spending hundreds of billions on AI data centers are using accounting trickery to reduce depreciation costs to the tune of a combined $176 billion from 2026 to 2028.

As glad as I am that Burry is shining a light on depreciation expenses, which many on Wall Street see as something to be adjusted away, his argument is wrong. When a company builds a data center from scratch it is buying assets like land, buildings, and equipment. Because these are large investments that tend to vary from year to year, GAAP accounting doesn’t expense it all at once on the income statement. Rather, the asset is depreciated over a set number of years, that is, its useful life. It’s those depreciation expenses that appear in a company’s income statement.

As Big Tech goes on its capital expenditure binge, most of the spending is tied to the equipment that sits inside data centers, much of it coming from Nvidia. Companies are depreciating the servers and networking equipment on a five-to-six-year schedule, up from four years in 2021.

The scheduled change makes for a big difference in annual expenses. Using a three-year depreciation schedule means that $300 million in data-center equipment gets expensed at $100 million a year. On a six-year schedule, it’s half that, saving $150 million in depreciation costs during the first three years. With most of Big Tech still in the early days of the AI capex binge, Burry’s short argument claims that profits are being inflated.

But these companies have good reasons for shifting depreciation. Burry seems to be assuming that each new generation of hardware obsoletes the previous one, but we have hard proof that that’s wrong. Nvidia’s A100 AI accelerators were first used in 2020 and are still in wide use on the major clouds.

Depreciation expenses will rise quickly, as we’ve noted in this column, and it will decrease profitability, but there is no accounting skulduggery as Burry’s short call suggests. This isn’t Enron or WorldCom.

Even still, Burry’s worries set the stage for CoreWeave’s earnings report that came late Monday. The numbers were strong and yet the stock was down 16% the next day.

CoreWeave is an AI pure-play “neocloud” whose entire business is renting out AI servers in the cloud. It has a particularly close relationship with Nvidia, which is a CoreWeave main supplier, investor, and customer.

While demand for the AI cloud remains insatiable, CoreWeave had a hiccup in its data-center buildout: A vendor for land, building, and power failed to meet a deadline. This affected one customer that will begin its CoreWeave contract in the first half of next year instead of the fourth quarter. Revenue growth will suffer in the next two quarters as a result, but then jump.

None of this was good news, but if investors were on board with the long-term prospects of the neocloud, the stock wouldn’t have swooned so hard. As a pure-play, CoreWeave stock has become a decent indicator of sentiment around the AI trade.

By the time AI investors learned on Tuesday that SoftBank was selling the rest of its Nvidia stock, they were on high alert. The headlines suggested the world’s most AI-focused firm was jumping ship on AI’s most important stock. But this is really just a rearrangement of the deck chairs.

Before the year is out, SoftBank has committed to investing another $22.5 billion in OpenAI, the most important private company in the world. There has been a lot of speculation around how SoftBank would fund the stake, which was revealed when the company reported its second-quarter earnings this past week. The company had sold shares of T-Mobile US, Deutsche Telekom, along with its entire Nvidia stake.

Nvidia dropped 3%. But the message here isn’t about Nvidia but rather that SoftBank CEO Masayoshi Son sees his next generational opportunity. SoftBank is still fully vested in the AI trade.

Created with Highcharts 9.0.1CoreWeaveCRWV / NasdaqSource: FactSet

Created with Highcharts 9.0.1April 2025Nov.255075100125150175$200

None of this is meant to dismiss investor anxiety around AI hype. Bubbles are psychological and behavioral events where investors all rush a single asset class, in this case AI data centers, even as the customers for these data centers aren’t yet seeing a return on investment. Sales at OpenAI are growing rapidly, but the company had at least $12.6 billion in losses-before-taxes in the third quarter, based on disclosures from Microsoft , which is a big investor.

Since 2024, venture-capital firms have poured $278 billion into AI companies according to Crunchbase, much of which is going to cloud bills. Private investors are the ones behind all the insatiable demand, and that is bubble-adjacent behavior. We also see other familiar signs in some of the financing deals announced in the past few months. What happens to the AI trade right now depends on the confidence of this relatively small group of venture capitalists, corporate CEOs, and private-equity executives.

But this week’s reactions were overdone. A bubble can go on for years before it pops, and none of the stories this week is evidence that we are getting any closer to the end.

Write to Adam Levine at adam.levine@barrons.com