Six more months of energy winter... Laying mines versus finding them... The real headwind for the stock market... The AI boom is falling apart... From tech adventure to smokestack hunting... There's a reckoning ahead...
No matter when the war ends, higher oil prices are here to stay...
Last week, the Washington Post spelled the problem out perfectly:
It could take six months to fully clear the Strait of Hormuz of mines deployed by the Iranian military, and any such operation is unlikely to be carried out until the U.S. war with Iran ends, the Pentagon has informed Congress – an assessment that means the conflict's economic impact could extend late into this year or beyond.
So even if the conflict were to end tomorrow, it could take another six months (at least) before oil flows freely through the Strait of Hormuz. It's like energy Groundhog Day. The little fellow saw his shadow and now we have at least six more months of energy winter.
You see, although the U.S. Navy has destroyed many of Iran's larger ships, the country still has plenty of small, fast boats to deploy mines in the strait.
Iran has reportedly deployed two basic types of mines: The Maham 3 is an anchored mine deployed in waters up to 328 feet deep, and the Maham 7 sits on the seafloor in shallower waters. And neither of these mines explode through contact. They're triggered by magnetic and acoustic sensors capable of detecting passing ships.
The impact of all those mines was expressed well by an April 9 Lloyd's List report... There are more than 600 commercial ships trapped in the Persian Gulf, including 325 oil tankers. The human toll is even more staggering. An April 29 report from Ynet Global estimates there are 20,000 sailors trapped on as many as 2,000 ships.
The ultimate problem is that laying mines is much easier than figuring out where they are...
Even though Iran left navigable shipping lanes unmined for ships prepared to pay a toll, it admits it doesn't know where all its mines have been laid. At least, that's what it claimed when President Donald Trump requested it allow more ships to pass through the strait.
Meanwhile, Brent crude, the global benchmark for oil, has been through a distinct "war on", "war off" cycle over the past month.
After surging to a four-year high of $118.35 per barrel on March 31, it quickly retreated to $90.38 by April 17, when folks were confident that Trump might soon end the conflict. When it became apparent that the term "ceasefire" doesn't mean everybody will stop shooting, Brent took off again, topping $118 per barrel. It recently traded around $114, signaling we're still in war-on mode.
But the verdict from the stock market is clear... $100-plus oil is no problem...
Yesterday, the Nasdaq Composite Index and S&P 500 Index reached new all-time highs.
The S&P 500 is now trading at 31 times earnings and 40.9 times cyclically adjusted price to earnings ("CAPE"). That's 8% shy of the highest level in history, going back to 1871.
The highest CAPE ratio ever was 44.2 in December 1999, a few months before the dot-com peak.
The Internet has been an incredible economic boon, creating massive wealth and new opportunities, including businesses now worth trillions of dollars.
And yet, that didn't keep the stock market from becoming insanely overvalued and risky – even for "no brainer" investments like fast-growing, cash-gushing "Internet plumbing" provider Cisco Systems (CSCO). Cisco was the No. 1 Internet-darling stock of the late 1990s. It became so overhyped and impossibly overvalued that it didn't eclipse its March 2000 high until December 2025.
We're seeing a similar situation play out today with the AI bubble. It will no doubt create trillions of dollars of business value in whole new industries, some of which probably don't even exist yet.
But just like with the Internet revolution, a lot of folks are buying a bunch of stocks that are going to get murdered once the bubble deflates, as they always do.
In short, the most likely headwind for U.S. stocks isn't oil – it's AI...
Right now, everybody is excited about AI's possibilities. The current AI market darlings are performing well. Only Meta Platforms (META) has failed to outperform the overall market since the March 30 bottom.
However, the biggest AI data-center builders' balance sheets continue to deteriorate. Nobody cares about it now, but I (Dan Ferris) suspect the day will come when that matters more than anything else.
As a group, Oracle (ORCL), Microsoft (MSFT), Alphabet (GOOGL), Meta, and Amazon (AMZN) had net cash of $116 billion at the end of 2021. Today, after borrowing hundreds of billions of dollars, these hyperscalers have net debt of $223 billion. And as you can see in the chart below, their borrowing levels have been rising rapidly since the end of 2024.
They're taking on debt to invest massive amounts of capital into AI. In 2023, their capital expenditures ("capex") were $149 billion as a group. Over the last 12 months ("LTM"), their capex has ballooned to $482 billion. It will easily exceed $700 billion by this time next year.
That's a lot of new debt for capex that won't make money over any reasonable time frame, if ever.
The AI data-center boom is already starting to fall apart...
As the Financial Times reported on April 17:
Delays to new US data centres threaten to slow the rollout of AI by the world's biggest tech companies, with almost 40 per cent of projects due this year at risk of falling behind schedule.
Major projects for Microsoft, OpenAI and other tech groups are likely to miss completion dates by more than three months, according to data shared with the [Financial Times] by SynMax, a satellite and AI analytics group.
Three months doesn't sound like a big deal. But I doubt folks are admitting how far behind they really are. SynMax tracks construction progress through satellite imagery. It estimates that more than 60% of projects scheduled for completion next year haven't even begun construction yet. For example, it's seeing limited progress on Oracle-, OpenAI-, and SB Energy-related projects in Texas. Yet all three companies claim to be on schedule.
The Financial Times cited "more than a dozen industry executives" complaining of "permitting hurdles and chronic shortages of labour, power and equipment."
The whole thing sounds exactly like the shale boom in North Dakota during the early 2010s, where oil-rig workers made six-figure salaries, fast-food places had trouble finding workers at $15 an hour, and Walmart paid nearly 2.5 times the minimum wage.
Data-center construction workers in Texas are moving from one OpenAI project to another in search of higher paychecks. That's right... OpenAI is competing with itself for construction workers.
Wes Cummins, the CEO of Applied Digital (APLD), which builds and operates data centers, told the Financial Times:
We're going to see a number of blow-ups and delays this year. My focus is on making sure it's not us.
Many of the new data centers are gigawatt-scale projects...
That means they're so big they'll require 1 gigawatt of electricity – enough to power more than 700,000 American homes. It's equal to 1.3 million horsepower, or the combined output of 2,000 Corvette Z06s. It's like dropping a small city on 500 to 800 acres of land.
I suspect many permitting delays are due to local opposition. The same folks who don't want a power plant in their backyards also don't want data centers there. But power is the ultimate data-center bottleneck.
There isn't enough power capacity for all the data centers planned and under construction. And even if you could build new power plants quickly enough, it's a seven-year wait to connect new plants to the transmission grid.
That's why hyperscalers are increasingly planning to build their own power plants. The overwhelming choice is natural gas turbines, but the big turbine builders have multiyear backlogs.
So data-center builders are getting creative. They're using turbines originally designed for airplanes and warships, mobile gas generators strapped to semitrucks, and refurbished turbines from industrial plants.
I thought the AI boom would be all about technology...
When I started researching AI last fall, I figured I'd wind up learning about cool tech companies doing amazing things...
There are buildings going up all over the country, filled with the latest, most powerful graphics processing units, running software that does the unimaginable.
But it turns out, all those buildings are in desperate need of construction workers and electricity, including backup power from diesel generators.
So instead of searching for the latest tech company, I've been hunting for smokestacks: companies that have been producing oil and gas, refined petroleum products, and chemicals – in some cases for more than 100 years.
I'm still bullish on AI...
The five hyperscalers I mentioned earlier all have cash-gushing businesses with big competitive advantages and high returns on capital without massive amounts of investment. And they're all betting big on the AI boom.
I believe this massive AI investment will yield incredible results – for consumers and professionals. Just like the Internet let a bunch of us make our living from anywhere we like and buy everything from fresh produce to custom-made suits, AI will change our day-to-day lives.
But I also have a feeling that hyperscalers will watch their balance sheets, cash flows, and hopes of an AI financial miracle deteriorate... and that the companies that make up 40% of the S&P 500 today will be replaced, perhaps with companies that mine copper, make diesel fuel, or some other essential product that can't be replaced by intelligent software code.
Maybe I'm premature in my worries (wouldn't be the first time!). Of the five hyperscalers I listed above, only Oracle is generating negative free cash flow today. It burned $25 billion of cash over the past 12 months. And it continues to spend more capital on data centers than it generates in operating cash flows. In other words, it's burning cash, not earning it.
In a recent interview for the Stansberry Investor Hour podcast (keep an eye on investorhour.com for the full episode), we spoke with veteran fund manager George Noble. George was once legendary investor Peter Lynch's assistant. He also oversaw the Fidelity Overseas Fund when it was the No. 1 fund in the country. So he knows what he's talking about. And he's no stranger to controversial opinions...
George spent a good part of our talk explaining why Tesla (TSLA) is worth roughly $25 to $54 per share, well below the recent price of around $395 per share. As we were wrapping up our talk, he also said he believes Oracle will go bankrupt.
The overall tenor of our conversation was that there's a reckoning ahead for folks who are paying way too much for companies making promises they won't be able to keep.
New 52-week highs (as of 4/30/26): ABB (ABBNY), Atlas Energy Solutions (AESI), Advanced Micro Devices (AMD), Amazon (AMZN), Alpha Architect 1-3 Month Box Fund (BOXX), Ciena (CIEN), Cisco Systems (CSCO), Deluxe (DLX), DXP Enterprises (DXPE), Emcor (EME), iShares MSCI Emerging Markets ex China Fund (EMXC), EnerSys (ENS), iShares MSCI South Korea Fund (EWY), Flex LNG (FLNG), Cambria Foreign Shareholder Yield Fund (FYLD), Alphabet (GOOGL), iShares Convertible Bond Fund (ICVT), Idex (IEX), KraneShares Bosera MSCI China A 50 Connect Index Fund (KBA), KraneShares Global Humanoid and Embodied Intelligence Index Fund (KOID), Keyence (KYCCF), Lumentum (LITE), Altria (MO), Nucor (NUE), Plains All American Pipeline (PAA), Pembina Pipeline (PBA), Invesco WilderHill Clean Energy Fund (PBW), Invesco High Yield Equity Dividend Achievers Fund (PEY), Invesco Oil & Gas Services Fund (PXJ), Ryder System (R), USCF SummerHaven Dynamic Commodity Strategy No K-1 Fund (SDCI), U.S. Global Sea to Sky Cargo Fund (SEA), State Street SPDR Portfolio S&P 500 Value Fund (SPYV), Tenaris (TS), Viper Energy (VNOM), and State Street SPDR S&P Semiconductor Fund (XSD).
In today's mailbag, a reader shares his take on Microsoft's spending plans, which we mentioned in yesterday's Digest... Do you have a comment or question? As always, e-mail us at feedback@stansberryresearch.com.
"Microsoft will spend $190 billion on capital spending. Their old run-rate was around $25-28 billion, so we can estimate their capex on data centers is around $165B. The buildings are depreciated over 30 years and the servers [and equipment] over no more than six years according to my research [Editor's note: Yes, generally true] – 4 or 5 years is more typical – but I will use 6 years...
"If we're generous and say that half the data center capex is on complex buildings, infrastructure, HVAC and power generation, then that's $80 billion, in round figures. That's $2.7B in annual depreciation... [But] then they have operation and maintenance costs which will run in the low percentages, plus enormous power costs... MSFT's annual depreciation in the prior fiscal year was $34B... This is a big nut to overcome... I'm not sure this will work out well at all." – Subscriber Mark P.
Good investing,
Dan Ferris
Medford, Oregon
May 1, 2026
