Written by Bryan Lutz, Editor at Dollarcollapse.com:
The S&P is at all-time highs…
While only a few AI-focused stocks drive the entire index up and to the right.
And meanwhile, the actual capital expenditure that’s supposed to justify the entire AI thesis is not getting built.
Of the 16 gigawatts of US data center capacity scheduled to come online in 2026, only 5 gigawatts are actually under construction.
That’s 31%…
Or, looked at the other way: 11 gigawatts of announced capacity has not broken ground, despite typical 12-to-18-month build times.
The thing is, there’s a bunch of steps that must take place before building can start. City council approvals, environmental checks, extended infrastructure plans. And that means, you can’t build it in time. So you announce it, book the press release, watch the stock pop, and quietly let it slip into 2027. Where the gap between announcement and action becomes even wider.
Now, Bank of America noticed something this week.
Bloomberg reports:
‘Bubble’ Signal Flagged by BofA as Tech Stocks, Volatility Soar
“A record-smashing rally in technology megacaps has rekindled bubble warnings at Bank of America Corp. To the firm’s strategists, a pattern similar to the dot-com bubble is playing out in the Nasdaq 100 Index, which has risen in tandem with the gauge’s measure of 10-day realized volatility for 14 of the 16 days through Wednesday. That’s a departure from a historical trend when price swings subside as shares rise.”
You see, prices and volatility both going up at the same time is the dot-com pattern.
Healthy bull markets compress volume as they climb. Bubbles expand volume on the way up because the market is paying for stories, not earnings.
BofA’s own economists were saying “no bubble yet” four months ago. Yet, their strategists are now saying the technical signal is here, which is the kind of internal contradiction that shows up at tops…
Here’s what the CapEx picture actually looks like, year by year. Announced versus actually under construction:
Look at 2027.
Twenty-five gigawatts announced. Ten under construction. The gap widens the further out you look, which is not a build cycle. It’s a press release cycle.
Like I was saying…
Earlier this year, Alphabet, Amazon, Meta, and Microsoft, committed $650 to $700 billion to AI CapEx, it depends on power infrastructure that doesn’t exist, transformers that ship from China, and electrical grids that need rate-payer-funded upgrades.
As a result, the rush is on to update the American electrical grid. US utilities are planning to spend $1.4 trillion over the next five years to power the AI buildout, per Yahoo Finance.
That spending requires regulated rate increases approved by state utility commissions. It requires federal loan guarantees. It requires the State to subsidize the grid that the ai hyperscalers depend on.
So, the CapEx isn’t private. It’s fiscally subsidized.
The bubble runs on the same printing press as everything else.
In fact, contrast that with Newmont last week.
Newmont reported $3.1 billion in free cash flow for Q1 2026. An all-time quarterly record. At an all-in sustaining cost of $1,029 per ounce, with gold at $4,580.
That’s a margin of roughly $3,550 per ounce of physical product they actually dig out of the ground and sell.
A 78% gross margin on gold.
Compare that to OpenAI, which Reuters reported is facing investor scrutiny over its $852 billion valuation while burning billions.
One company is profitable, hated, and cheap.
The other is unprofitable, loved, and at all-time highs.
And the chart that tells you the trade has been working for two years:
Gold miners up 190% since January 2024. Mag 7 up 115%.
The miners outperformed and are still cheap.
And Mag 7 is still running on stories, while the miners are now running on cash.
Average joe’s like you and I have been told for two years that the AI thesis is the only thesis that matters. That the index is held up by capex that will eventually arrive. That valuations are justified by future productivity gains nobody has yet measured.
Meanwhile, half the data centers aren’t being built. The biggest pure-play AI company is burning through cash. And the BofA strategists are quietly drawing dot-com comparisons in the same week the firm’s economists are still bullish.
So here’s the question for this week.
If the AI buildout is the only thing holding up the index, and the buildout is contingent on State-subsidized grid capacity that takes years to permit and parts that ship from a country we’re in a trade war with…
What exactly are the Mag 7 multiples pricing in?
One of these (from the GDX and Mag 7) companies is making money.
The other is making promises.
