The Looming Tower: The Unsexy Backbone of the AI Boom
By Bryan J. Kaus
Most people think the AI boom is a software story.
It’s not.
It’s a physical buildout story — one that runs through data centers, power generation, transmission, fuel deliverability, permitting, and local infrastructure. And the part that scales slowest becomes the governor on everything else.
That’s the risk, the opportunity, and the hidden market structure shift happening in front of us.
The Through Line Most People Miss
Data centers can be planned, financed, and poured quickly compared to the grid and the equipment that feeds them.
But what’s hitting the system isn’t incremental. It’s a step-change in load expectations.
One widely-cited synthesis of U.S. utility load forecasts projects ~166 GW of peak load growth by 2030, with roughly ~90 GW tied to data centers. That’s not a rounding error — that’s a new era.
And it gets more complicated: even the forecasters are warning that some projections may be overstated due to unrealistic assumptions about utilization and load factors — meaning the shape of demand might differ from the headline number.
That combination — big demand + forecast uncertainty — is exactly how you get overbuild in some pockets and scarcity in others.
The Load Stack
If gasoline prices are a “stack,” AI load is a stack too.
Here’s what actually has to clear for an AI/data center boom to translate into durable cash flows across the value chain:
Permitted sites + community license
Interconnection rights (and realistic upgrade costs)
Transmission availability (and who pays for it)
Grid equipment (transformers, switchgear, breakers — the quiet bottleneck)
Firm generation (not just nameplate capacity)
Fuel deliverability (gas pipelines, compression, storage — and winter performance)
Water + cooling solutions (in the places where water is already political)
Any one of those can become the constraint. But in the U.S. right now, two keep showing up: interconnection and equipment.
Interconnection: the slow choke point
Interconnection is where great spreadsheets go to die.
In the latest “Queued Up” work from Lawrence Berkeley National Laboratory, the typical project built in 2023 took nearly five years from interconnection request to commercial operations.
That single statistic should reframe how you think about “near-term” power solutions.
You can want a data center online in 18–24 months. The grid may not agree.
Turbines: the supply chain is speaking
This isn’t theoretical demand anymore. The OEM queues are telling you what’s real.
In GE Vernova’s January 28, 2026 earnings webcast transcript, the company noted Gas Power equipment backlog and slot reservations rising from 62 GW to 83 GW sequentially, and management expects to reach ~100 GW under contract in 2026.
That’s what bottlenecks look like in plain English: slots become the product.
Why Natural Gas Is Back in the Driver’s Seat
If you need power that is:
dispatchable
financeable
available at scale
compatible with data center uptime requirements
…you end up back at gas-fired generation, at least in the current build cycle.
But here’s the second-order reality: “sea of gas” doesn’t automatically mean “gas where you need it, when you need it.”
The reliability community has been blunt about this. The North American Electric Reliability Corporation winter assessment warns that the performance of natural gas production and supply infrastructure during peak winter conditions can significantly affect bulk power system reliability.
Translation: deliverability matters. Winter performance matters. The weakest links show up under stress — and stress is not hypothetical.
The Precarious Machine
When you tie these together, you get a system with a very specific failure mode:
Data centers scale fast because capital is abundant and incentives are aggressive.
Power generation scales slower because equipment and permitting are real.
Transmission scales slowest because it’s a coordination problem disguised as engineering.
Midstream responds to “obvious demand” signals, but demand can be lumpy, delayed, or regionally stranded.
That’s how you get overbuild and underbuild at the same time.
And over a cycle, that creates exactly what seasoned capital allocators recognize:
stranded commitments
empty capacity
renegotiated contracts
force majeure disputes
margin compression
debt metrics tightening at the worst moment
opportunistic consolidation when weaker players break
If you operate inside one segment, this can feel like a random weather event.
From the “whole-board” view, it’s a predictable outcome of mismatched build tempos.
Where the Opportunity Actually Is
There’s a tendency to chase the “sexy” layer (chips, AI models, hyperscaler headlines).
But the durable advantage is often in the boring connective tissue:
grid-enabling equipment and services
permitting + interconnection navigation
dispatchable generation development + repowers + uprates
fuel deliverability solutions
midstream expansions that solve specific constraints — not vanity buildouts
In other words: the opportunities live where the stack is tight.
And there’s a very practical strategic logic for midstream in particular:
“Shortline” thinking for energy infrastructure
In the 1800s, the winners weren’t only the giant railroads. Plenty of fortunes were made in local and regional “shortline” buildouts that stitched supply to main corridors — then got rolled up.
A modern analog exists in the last-mile and constraint-clearing layers of energy infrastructure:
targeted lateral pipelines
compression additions
storage optimization
interconnects that unlock a constrained market pocket
paired generation + fuel solutions that de-risk deliverability
This is where specialists can build, prove cash flow, and sell to a scale provider — or roll into a platform.
It’s not glamorous. It’s how systems get built.
A Preview of the Operator’s Playbook
This is where leadership separates from optimism.
The question isn’t “Will something break?”
It’s “Where does it break first — and how exposed am I when it does?”
Here’s the advisory-grade lens that matters:
1) Treat demand as probabilistic, not binary
Contracts, underwriting, and capex should be built around scenarios:
base demand
delayed demand
partial demand
demand migration (same customers, different nodes)
2) Build optionality into your capital plan
Stage gates. Modular expansion. Convertible designs. Step-in rights.
3) Use contract structure as a shock absorber
(Yes, capture these terms — they’re the difference between compounding and litigation.)
take-or-pay and/or minimum volume commitments
step-outs and re-pricing triggers
escalation clauses tied to inputs that actually move
interconnection/upgrade pass-through logic
curtailment language that’s realistic, not aspirational
credit protections that assume a downturn happens before the best case is realized
4) Share risk like an adult
Joint ventures aren’t just for funding — they’re for risk partitioning.
Smart structures often include:
clear governance
ROFR/ROFO mechanics
buy-sell provisions
paths to spin-outs or drop-downs
the ability to capture upside without single-project fatal exposure
When the cycle turns, the people who survive are the people who engineered survivability up front.
What to Watch
If you want to track this market without getting lost in hype, keep your eyes on the stack:
Utility load forecast revisions (and the assumptions behind them)
Interconnection timelines + upgrade cost volatility
Transformer / switchgear lead times (this is the quiet limiter)
Gas turbine backlog + slot availability
Winter reliability notes on gas deliverability
Midstream contract terms tightening (a sign of a smarter market)
Where incentives are being offered vs. where power can actually clear
M&A patterns (distressed cleanup vs. premium platforms)
Debt covenant pressure in capex-heavy names
Regional basis signals (where the system is screaming “constraint”)
The Point Taken:
AI is not only a software boom — it’s a load shock.
Data centers scale faster than the grid can respond, which guarantees mismatches.
Those mismatches create both outsized opportunity and predictable blow-ups.
The winners will be the operators and investors who understand the full stack — and structure capex, contracts, and partnerships so they can survive the cycle and capture the upside.
If you’re building, financing, or operating inside any part of this stack (data centers, power, midstream, industrials), I’ve written a deeper plan that lays out the full framework: where the choke points form, how overbuild cascades, and how to structure projects so you don’t get trapped when the cycle turns. Fortune favors the bold. Sustained value is found through agility.