Published on Nov 6, 2026
This article was written by Adem Tumerkan
Content Strategist, Dunham | 2025 ThinkAdvisor Luminaries Award Winner | 2026 Wealthies Finalist, Thought Leader of the Year | Macroeconomics, markets, geopolitics & global trends
Subscribe to get the latest insights delivered straight to your inbox.
We care about your data in our privacy policy
The AI infrastructure boom faces three potential constraints: financing costs, access to electricity, and political opposition to data centers. Together, these pressures could delay projects and weaken investment returns even if demand for AI keeps growing. This article applies Clausewitz’s “culminating point” to the buildout - the possibility that expansion outruns the resources needed to sustain it. Whether that point has arrived remains an open question.
A few years ago, an old mentor of mine insisted I read Michael Handel’s Masters of War, a study of history’s great military strategists.
I remember thinking, “Of all the books, why this? What does this even have to do with markets?”
Well, I read it. It was dense, but I can see why he pushed it on me.
There were plenty of nuggets in there, but one term really stuck with me. . .
It’s a military idea, but it explains a lot outside of war too – especially in markets.
It’s called the culminating point.
The gist is that armies don’t always lose at the front – but when they outrun their supply lines.
Carl von Clausewitz, a Prussian officer, crossed into Russia in 1812 to join the Russian army just as Napoleon was massing troops to invade. He watched the greatest army in Europe march toward Moscow and come apart on the way back1. It became the textbook case for his idea.
Which brings me to AI.
The AI buildout looks a lot like an offensive that’s starting to stretch. Its supply lines are cheap credit, grid power and public consent. Lately, all three have gotten tighter.
I think the AI buildout may be closing in on its own culminating point.
From here, bond investors, grid operators and voters could get more say over how fast this goes than Big Tech does – and that’s a big deal for a market pumped up (borderline dependent) on AI.
Put simply, the culminating point is the moment an attacker’s advantage runs out, when stretched supply lines and a stiffening defense hand momentum to the other side.
Clausewitz laid it out in On War, first published in 1832.
The idea was pretty simple - an army wins a battle and keeps moving forward. But every single mile takes it further from its own barracks and supplies while the retreating defender falls back closer to his supply lines and reinforcements. Worse, the attacker also has to leave troops behind to hold roads, towns, etc., which thins the force at the front.
He compared a conquering army to a lamp. As the oil in the reservoir sinks further from the flame, the light shrinks until it finally goes out.
Meaning - by the time the light dims, the oil has been sinking for miles.
The AI buildout now drives a large share of U.S. earnings growth, GDP growth and household wealth, which makes any slowdown in its supply lines an economic-wide problem.
Right now, the offensive was the capex.
Goldman expects hyperscaler capital spending of roughly $800 billion this year – hitting $ 1.4 trillion in 2028. So while it’s trending higher, the growth of this capex is slowing significantly percentage-wise.
Meanwhile, almost half of the S&P 5002’s earnings growth this year traces back to this AI investment and buildout.
And finally, Moody's estimates3 data center and computer investment added about 1 percentage point to U.S. GDP growth (that's roughly a third to a half of recent growth).
And that’s before the ripple effects. The AI-fueled stock rally helped add 736,000 new millionaires in 2025, bringing the total to 8.7 million, or roughly 2.5% of the population4
The point is, all this AI spending has fueled corporate earnings, economic growth, and household wealth at a staggering pace.
AI borrowing is flooding bond markets with long-term debt, and that added supply is helping push yields higher - first on corporate bonds and increasingly on Treasuries.
In the early days, cheap capital fueled the AI buildout.
The hyperscalers were making so much cash that they could pay for most of it themselves.
But that’s changed. Now they're turning to debt for the next leg of growth.
For instance, the largest hyperscalers sold about ~200 billion of bonds in 2026 through early July, already 79% more than they issued in all of 2025 and making up about 9% of all investment-grade debt issued. And Goldman Sachs expects5 the five biggest to issue around $250 billion this year and $400 billion in 2027.
Here’s a simple way to see why that matters.
Say an investor has $1,000. They can lend it to the U.S. government by buying a Treasury (which is considered risk-free). Or they can lend it to a company by buying a corporate bond. Companies can default, so they pay a little extra yield to get that $1,000.
That extra is called the spread.
Usually, when companies flood the market with debt, they have to raise that sweetener to find buyers. This time, investors wanted AI debt so badly that spreads barely moved. So the Treasury had to compete instead, and Treasury yields rose to keep buyers coming.
Even the Fed has noticed. Fed Chair Kevin Warsh recently named hyperscaler borrowing as one of three reasons yields have surged, with the 10-year up nearly a full percentage point over the past year.
But PIMCO had a different take. It thinks AI pushes rates up more indirectly, by soaking up power, steel and workers, which keeps inflation sticky6.
Either way, it loops back on AI.
A corporate bond’s yield is the Treasury yield plus the spread. So even with spreads flat, AI’s own borrowing costs climb right along with Treasuries.
So, the attacker’s cheap credit supply line is starting to come apart.
The power grid has become the slowest supply line in the AI buildout - with grid capacity and connection equipment lagging data center demand in many regions.
The point is, if credit can be raised in a day. Building new power takes years.
Much of the grid is already decades old – with research showing that roughly 70% of transmission lines and large power transformers were more than 25 years old.
Now, data centers need huge amounts of power around the clock.
For instance, MISO - which operates across much of the Midwest and parts of the South - says large users want connections within 18–36 months7, far faster than new generation and transmission can typically be built.
And this spread between demand and supply is causing prices to surge.
PJM – the nation's largest grid operator – saw capacity prices jumped 833% between the 2024–25 and 2025–26 delivery years (meaning the cost of keeping power plants available for peak demand surged, adding pressure to electric bills). And even at those higher prices, PJM didn’t secure enough power capacity to meet its target for keeping the lights on during periods of peak demand.
Texas faces a similar issue.
ERCOT reported roughly 474 gigawatts of large-load connection requests as of June 2026, about 90% from data centers8. Now, those are just proposed projects, but they show the scale of the rush for power.
Highlighting this - according to Bloomberg - data centers in the US will account for about 20% of the nation’s electricity consumption in 2035 vs. 5.9% today.
An army can’t march faster than its wagons.
If credit and power were the stretched supply lines, Tuesday’s “blue wave” - which gave Democrats control of the House and Senate - brought a counterpunch. Opposition to data centers featured in several winning campaigns, putting the third supply line (public consent) under more pressure.
Most surprisingly was that in Texas and Georgia - Democrats won governorships that their party hasn't held in decades9.
Gina Hinojosa unseated Abbott in Texas, a state that was expected to host ~41% of new U.S. data center capacity by 2028. Abbott had already hit the brakes on new grid hookups, but he lost anyway. In Georgia, Keisha Lance Bottoms is the state CBRE calls home to the country’s busiest data center construction market. At least 14 Ohio municipalities have also restricted data center development10.
Of course, there are many reasons for an election – but data people not wanting data centers in their backyards or the power costs that get passed onto them was a big part of it.
AI data center debt gets riskier when power or permits arrive late, because interest keeps piling up before any lease revenue starts.
This is where all three issues meet.
A data center builder borrows today, and interest starts stacking up on a fixed schedule. But revenue only arrives once the site has power and the tenant’s lease starts. Thus, if power doesn’t come, interest keeps compounding regardless.
Said another way, it’s not like bond payments stop just because a data center calls and says, “Oh geeze, we can’t get power right now. Can we hit pause?”
Thus, the cushion – the gap between when bond payments start and revenue comes in – is very thin.
To put this into perspective, a recent piece published by Brookings11 estimates that the five biggest hyperscalers will spend about $800 billion on buildings, equipment and other major investments in 2026 - roughly $93 billion more than their businesses generate in cash. That leaves a gap to fill with borrowing, outside investors or existing cash reserves.
And making the buildout pay off will take enormous revenue. The same analysis estimates the U.S. AI buildout would need about $3.7 trillion in annual revenue by 2032 to deliver investors a 10% return (assuming half of every revenue dollar remains after operating expenses).
Every month power is delayed shows up late is another month of interest with no rent coming in.
Even with headwinds, the AI offensive is moving ahead.
Revenue at one of the largest AI labs jumped 47% in the third quarter, and plenty of analysts argue that fewer U.S. data centers wouldn’t mean less demand for AI (if anything, it may help returns due to limiting overcapacity - which I've outlined before as a major threat to the AI boom).
Meanwhile, a slowdown could also pull yields down. Goldman’s Jan Hatzius expects AI’s boost to growth to crest in late 2026 or early 2027. On Wednesday, AI chip stocks fell in Asia, and bond yields fell with them. But deficits and fuel prices would still be there, so yields may not fall as far as the AI angle suggests.
But for now, it looks like the supply lines are stretching faster than the front is moving.
AI can win this war and still have to slow its march.
Nobody knows exactly where the culminating point sits. Clausewitz didn’t claim to either.
But the fragility - and uncertainty – are rising.
As always, time will tell.
This communication is general in nature and provided for educational and informational purposes only. It should not be considered or relied upon as legal, tax or investment advice or an investment recommendation, or as a substitute for legal or tax counsel. Any investment products or services named herein are for illustrative purposes only and should not be considered an offer to buy or sell, or an investment recommendation for, any specific security, strategy or investment product or service. Always consult a qualified professional or your own independent financial professional for personalized advice or investment recommendations tailored to your specific goals, individual situation, and risk tolerance. All examples are hypothetical and are for illustrative purposes only.
Information contained in the materials included is believed to be from reliable sources, but no representations or guarantees are made as to the accuracy or completeness of information. This document is provided for information purposes only and should not be considered as investment advice.
Dunham & Associates Investment Counsel, Inc. is a Registered Investment Adviser and Broker/Dealer. Member FINRA/SIPC. Advisory services and securities offered through Dunham & Associates Investment Counsel, Inc.
