Cracks Appear in the AI Trade, But the Bubble Hasn't Burst Yet

Timing matters in financial journalism as much as in investing. Barron's proved that in March 2000, when its cover story "Burning Up" pointed out that many dotcom darlings would run out of cash before year end. It ran 10 days after the Nasdaq's peak. Markets dried up, valuations collapsed, and it took 15 years for the index to reclaim that high.

With the Nasdaq and S&P 500 setting fresh records this week, the instinct to hunt for cracks in the foundation is understandable. There is no shortage of candidates.

Firmus Technologies, an Australian data centre operator backed by Nvidia and Blackstone, had hoped to pull off the country's largest IPO in three decades. Executives pitched a novel metric, EV+1/EBIT+2, comparing enterprise value 12 months out to earnings before interest and tax two years hence. The pitch was meant to help investors look past heavy upfront spending toward future payoff. It failed. Firmus pulled its $31 billion listing on Friday, citing "recent market volatility and prevailing market conditions."

Then came the OpenAI revision. The Financial Times reported Thursday that the company's annualised revenue run rate in September was about $50 billion, some $20 billion below figures previously cited by the FT and others. The number itself is less important than what it reveals: multiplying a single month's revenue by 12 is a crude way to size up a company's financial health. Still, the disclosure triggered a broad selloff in listed tech names.

Why the Bubble Hasn't Popped

Neither setback is fatal on its own. Other data centre operators are generating solid returns on chips and energy spending. SpaceX's chief financial officer has said the company should recoup its initial outlay in under a year, according to Breakingviews reporting. OpenAI's reported trillion-dollar-plus valuation rests less on last month's revenue than on what the company might generate two or three years from now. Markets have also overreacted before and recovered. The brief selloff triggered by China's DeepSeek model release in early 2025 is a recent example.

What has genuinely shifted is the cost of capital. The yield on 10-year US Treasuries, the global benchmark for borrowing costs, has climbed roughly a percentage point since late April. That changes the math for debt-heavy data centre buildouts in a way that a single bad headline cannot. Anthropic's planned IPO, targeted for before the end of November according to Bloomberg, will be the next real test of investor confidence in AI valuations. Only in hindsight will anyone be able to point to the article that marked the top of this cycle, if indeed this is one.

Banks face their own version of disruption. The "SaaSpocalypse" hit software stocks when AI coding tools threatened the old subscription model. Now digital assistants like Meta's Muse could nudge previously complacent depositors to demand higher rates on savings, squeezing net interest margins. Stephen Gandel ran the numbers on the $3.8 trillion sitting in US savings accounts. Lifting the average rate from 1.9% to 4% would cost lenders roughly $79 billion a year in lost income. Call it NIMageddon.

Elsewhere, the business world has its own reckonings. The Manchester City sham-contracts scandal, worth $1.1 billion over nearly a decade, has implications well beyond English football. It raises hard questions about whether a level playing field is even possible when financial stakes in sports franchises have grown this large, a point debated on Breakingviews' Viewsroom video series this week.

There is also a broader question about whether markets are becoming less rational, or simply more visibly idiosyncratic. Academic Alex Edmans, author of "The Madness of Markets," argued on The Big View that the old rules of investing are shifting in ways that deserve scrutiny rather than dismissal.

Finally, there is BYD's ambition to overtake Toyota as the world's largest carmaker by 2030. Toyota's path from Japanese upstart to global leader offers a template, but BYD's road looks rougher. Stagnating domestic sales, export barriers, hostile foreign governments and brutal competition at home all stand in the way. The comparison is instructive, but execution will matter more than ambition.

What Investors Should Watch

None of this adds up to a verdict on whether AI valuations are in a bubble. What it does show is that the market's tolerance for unconventional metrics and crude revenue extrapolations is thinning. Firmus tried to sell investors on a two-year-forward earnings multiple and found no buyers. OpenAI's revenue revision, even if methodologically unsurprising, was enough to spook tech stocks broadly.

The real pressure point is rates, not headlines. A one-point rise in the 10-year Treasury yield since April has quietly raised the hurdle rate for every debt-financed data centre project in the pipeline. That is a slower-moving story than a cancelled IPO, but it is the one that matters most for long-term AI capital allocation.

Anthropic's IPO will be the next meaningful data point. If it prices well and trades steadily, the market's appetite for AI exposure remains intact despite the yield backdrop. If it struggles, expect the Firmus episode to look less like an isolated misstep and more like an early warning. Investors should also watch bank margins closely as deposit competition intensifies, and keep an eye on BYD's execution against its 2030 target as a read on how global industrial competition is evolving outside the AI narrative entirely.