 |
| By Michael A. Robinson |
We’ve reached a tipping point. And it’s happening a whole lot faster than anyone expected.
Cloudflare’s (NET) CEO Matthew Prince says AI agents now surpass human traffic online.
In fact, it’s probably been the case for a couple of months.
In other words, AI continues to be the fastest adopted tech in history, and there’s no end in sight.
If anything, the use of AI agents to “think” and act on our behalf is only getting started as agent abilities grow by the day.
But even though we’re still in the early stages, AI’s bottlenecks are stacking up, and fast.
And that spells opportunity …
The Hardware Bottleneck
Take AI chips.
Despite investing a record $56 billion this year in new capacity, demand for the most advanced chips from the world’s premier chipmaker, Taiwan Semiconductor (TSM), will likely be 25% to 30% greater than supply.
And much of Samsung’s smaller share of AI chips are sold out this year.
Then there’s the packaging. TSM's advanced packaging lines are booked solid, with lead times north of 50 weeks.
Memory is just as tight. Micron’s (MU) high-bandwidth memory is sold out through 2026.
Of course, scarcity has a price.
Apple (AAPL) and Nvidia (NVDA) reportedly shell out around $30,000 for a single 2nm wafer now — a premium that crushes the smaller players who can't pay up.
Without a step-change in capacity, the AI hardware supply chain will stay tight for the next several years.
Power Play
Now let’s move to the one that hits a lot closer to home … electricity.
By 2030, data centers will need $6.7 trillion in global investment to meet demand for computing power, according to McKinsey.
And simply plugging into the grid no longer looks like a feasible option in the U.S.
Bloomberg just reported that the feds are floating a breakup of the PJM Interconnection, the largest U.S. grid operator that covers 13 states, from Illinois to New Jersey.
The reason? AI data center load is straining supply and spiking consumer bills (buckle up for a wild summer!).
Prices increased from $28.92 MW/day during 2024-2025 to $329.17 MW/day for 2026-2027.
That’s a 1,038% increase hitting the price cap two auctions in a row.
Data centers led to 63% of the 2025-2026 auction increase, or $9.3 billion in added costs for 67 million ratepayers.
Families face a projected $70/month increase in their energy bills in two years.
All told, costs could rise as high as $163 billion through 2033.
For the first time in its history, PJM’s capacity auction failed to meet demand in December by 5.2%.
It’s why states are moving to implement rate reforms, tax incentive rollbacks and construction moratoriums that could upend how data centers operate.
And the politics are turning nasty.
The governors of Pennsylvania and Virginia are threatening to pull their states from the PJM.
American Electric Power is weighing its own exit, as is the state of Maryland.
FERC (the federal energy regulator) Chair Laura Swett calls it "too big to function."
But voters aren't just mad about the high bills.
A recent Gallup poll found 71% oppose a data center nearby, compared to 53% who oppose a nuclear plant.
Going Behind the Meter
The smart money is already moving on from waiting on the grid to update.
Instead, they’re going “behind the meter” — meaning they’re building power right at the data center and skipping the public utility altogether.
Waiting on a grid hookup to the PJM, for instance, can take up to 12 years.
Combining on-site small modular reactors (SMRs) with natural gas generators can give data centers power in 24 months.
It’s why every major hyperscaler has signed a nuclear deal over the past year.
Meta (META) has signed deals to get up to 6.6 gigawatts of nuclear power over the next 20 years — one of the largest commercial nuclear deals in American history.
Amazon (AMZN) shelled out over $20 billion for a data campus in Pennsylvania. And Microsoft (MSFT) is restarting Three Mile Island.
Finding the Power Players
If you think you’re late to the party, think again …
Utilities with both gas turbines behind the meter projects and SMR roadmaps offer a diversified exposure to the growth ahead, plus the comfort of a steady dividend.
Dominion Energy is one example of a picks-and-shovels play that fits the bill with on-site gas projects and is considering nuclear deals.
NextEra Energy (NEE) recently announced that it plans to merge with Dominion, creating an even larger utility angle for this bottleneck.
Both companies have a long history of paying solid dividends. So, even though we don’t know what those will look like right now, the new combined company should be on your income radar.
Midstream gas provider Kinder Morgan (KMI) is partnering with hyperscalers on micro-grid gas projects and has a high Weiss rating, with a 3.71% forward dividend yield.
Engie SA (ENGIY) is a Weiss “Buy”-rated Europe-based power player with gas and nuclear projects in the works. It pays close to a 4% dividend yield.
Meanwhile, equipment makers like Siemens (SIEGY) can capture the high-margin aftermarket stream once plants are online, offering a different risk/return profile than regulated utilities.
The firm is also a “Buy” right now.
While it doesn’t pay quite as large of a dividend as the others, it gives you more international exposure.
Of course, if you are looking for more income from your investments, you’ll want to see what we’re working on right now.
Dr. Martin Weiss and income-expert Nilus Mattive are putting together a special event on Tuesday, June 30 at 2 p.m. Eastern.
You can grab the link to that here.
In short, you don’t need to own Nvidia and the other likely AI companies to profit from the bottlenecks their technology is creating.
And you can do a heck of a lot better than those tech giants if you want to also collect dividends from your portfolio.
Best,
Michael A. Robinson
1 https://x.com/eastdakota/status/2062212701414187452
2 https://www.utilitydive.com/news/pjm-ferc-swett-capacity-governance-data-center/820085/
