AI Infrastructure Stocks Shed $4.2 Trillion in Market Cap Since June
nashnova research
Global AI-infrastructure-linked companies have shed $4.2 trillion in market value since June, erasing nearly half of the prior 12-month rally — a systematic repricing of the "everything is a growth stock" narrative.
$4.2 trillion gone — what happened?
Global manufacturing, utility, and mining stocks have lost a combined $4.2 trillion since June, wiping out nearly half of the gains built over the previous 12 months.
This means → the sell-off is not about one company blowing up. Investors are collectively re-questioning the AI "picks-and-shovels" thesis — buying the firms that build AI's infrastructure instead of AI itself.
Per the Financial Times, the more closely a company was linked to data-center construction in media coverage, the harder it fell. Industrials are now the second-worst S&P 500 sector this summer, behind only tech.
Which heavyweights got hit the hardest?
Caterpillar, a key supplier of gas-fired generators for data centers, has lost $21.4 billion in market cap since its all-time high on June 30 — a 44% drop, with a 30%+ decline over six months.
Nuclear-power startup Oklo plunged 49% despite signing a power-supply contract with Meta in January. In plain terms = even a big-name contract could not shield it from a sector-wide crisis of confidence.
HVAC maker Comfort Systems and fiber-optic manufacturer Corning each lost roughly half their market cap over the same period.
Why the sell-off? Are the businesses actually deteriorating?
Deutsche Bank senior economist Brett Ryan noted that large pools of institutional capital had "piled in blindly," likely "inflating overly optimistic valuations" at some firms.
Venture capitalist and MIT researcher Paul Kedrosky was blunter: treating traditional industrial companies like semiconductor stocks was itself the mistake — "the 'everything is a growth stock' line was everywhere in hedge-fund and family-office conversations, and of course it was absurd."
This reflects a valuation problem, not an operational one. The market assigned these companies multiples that belong to tech, not industrials — and is now correcting.
But aren't the order books still strong?
Caterpillar reported a record $78 billion order backlog in Q3, with power-generation sales up 75% year-over-year. Siemens's data-center electrical-equipment backlog grew 60% in 2026 to €7 billion. Eaton's equivalent orders doubled year-over-year.
Dell'Oro Group analyst Alex Cordovil said backlogs are at historic highs, meaning revenue growth is "essentially locked in through at least 2027."
Kedrosky, however, is skeptical. He flagged "double and even triple ordering" — customers placing excess orders to secure queue positions amid perceived shortages. In plain terms = the headline backlog may overstate real demand.
What does the political and regulatory picture add?
After the U.S. midterm elections, ballot measures restricting data-center construction won broad voter support in several localities, clouding the regulatory outlook.
Citi Research's post-election note concluded that tighter regulation would add volatility but that a broad construction ban remains unlikely.
Constellation Energy CEO Joseph Dominguez called the market's post-election reaction an "overreaction"; Ryan argued that even localized bans would simply redirect construction to more permissive regions. This means → regulation is more likely to redistribute where data centers get built than to shrink total demand.
What is this correction really about?
This $4 trillion-plus wipeout is, at its core, a systematic repricing of the "everything is a growth stock" narrative applied to AI infrastructure.
Ryan's observation that equities are "catching up to the signal credit markets sent long ago" highlights the timing gap: bond investors spotted the risk first; equity investors are only now following.
The key test ahead: whether record backlogs convert into real demand — or mask a double-ordering bubble. That answer will determine whether this correction is halftime or the final whistle.
市场有风险,内容仅供研究参考,不构成投资建议。
