The Pareto Investor

The Pareto Investor

Nuclear Is The New Oil: AI’s Power Bottleneck Trade

Spoiler: It's Not Software Or Semiconductors. Sam Altman Needs Half Of America's Power By 2033. These Are The Only Companies That Can Deliver It!

The Pareto Investor's avatar
The Pareto Investor
Sep 19, 2025
∙ Paid
Logarithmic scale chart displaying OpenAI's operational energy capacity trajectory from beginning of 2025 at 0.23 gigawatts through end of 2025 at 2 gigawatts (9x growth), projecting to 250 gigawatts by 2033 with 1.8x annual growth rate. Right sidebar compares this to country energy capacities: United States 488 GW, India 223 GW, Germany 57.7 GW, United Kingdom 35.3 GW, Singapore 6.55 GW, Ireland 3.94 GW, Iceland 2.30 GW, and Malta 0.34 GW. Quote from Sam Altman via Slack included
OpenAI’s audacious infrastructure buildout: Starting at 230 megawatts in 2025, the company projects reaching 2 GW by year-end and scaling to 250 gigawatts by 2033—a capacity rivaling the United States’ 488 GW and surpassing India’s 223 GW. This 1.8x annual growth trajectory represents one of the most aggressive infrastructure expansions in corporate history.

Dear investors,

Sam Altman just dropped the most expensive prediction in history.

By 2033, OpenAI alone is projected to consume 250 gigawatts of energy—more than the entire country of India uses today.

In a leaked internal memo (reported by Fortune and others in Nov 2025), OpenAI CEO Sam Altman set an extremely ambitious long-term goal:

“Build or secure 250 gigawatts (GW) of dedicated power capacity by 2033 just for OpenAI’s AI training and inference needs.”

How much power is 250 GW really?

250 GW = 250,000 megawatts (MW) = 250,000,000 kilowatts (kW)

For comparison:

  • The largest nuclear power plant in the U.S. (Palo Verde) produces ~4 GW

  • The entire United Kingdom’s average electricity demand is ~30–35 GW

  • All of Google’s global data centers combined use roughly 15–20 GW (2024 estimate)

OpenAI alone wants the equivalent of 7–8 entire United Kingdoms’ worth of electricity, or roughly 15% of the entire United States’ total generating capacity, dedicated only to its AI clusters by 2033.

To manage this, OpenAI would need to buy ~30 million high-end GPUs every single year and run them 24/7/365.

Rough math:

  • Assume future GPUs in 2030 are 5–10× more power-efficient than 2025 H100s but models are 100–1000× bigger → net ~50–200× more total compute needed.

  • 250 GW of power ÷ ~800 W per future GPU ≈ 312 million GPUs running simultaneously

  • GPUs used for training burn out or become obsolete fast when run flat-out (typical lifetime 2–4 years in AI training)

  • So every year you have to replace ~1/3 of them → 100+ million replacements + growth → easily 30–50 million new GPUs purchased per year

Nvidia’s entire production of H100/H200/B100/B200 GPUs in 2024–2025 is estimated at only ~3–5 million units total worldwide. So OpenAI alone would need 6–10× Nvidia’s entire current output every year!

And here’s the kicker:

OpenAI will likely be just one of a dozen companies requiring that much power.

This isn’t about semiconductors anymore. The real AI supercycle is happening in energy infrastructure, and most investors are looking in the wrong direction.

The artificial intelligence revolution has captivated Wall Street with semiconductor stocks and software platforms, but the most explosive investment opportunity may be hiding in plain sight.​

While everyone chases the latest AI model or chip manufacturer, a parallel energy crisis is unfolding that will reshape the global economy and create generational wealth for investors who position themselves correctly

If you want to follow along with real money, here's how to copy my portfolio in 5 minutes.

The same AI boom that needs all this power is about to move into factories and homes, and I show who profits when robots leave the lab in Humanoid Robots Are Coming.


Join free


OpenAI Math That Breaks Everything

Sam Altman’s internal targets are physically impossible without unprecedented infrastructure spending.

OpenAI CEO has floated internal targets of 250 gigawatts of computing capacity by 2033—roughly half of America’s power demand.​

To put this in perspective:

If OpenAI achieved this capacity today, it would rank as the seventh-largest country in the world by electricity production capacity, ahead of France, South Korea, Brazil, and Saudi Arabia.​

And OpenAI isn’t alone in this power-hungry race.​

Citi analysts estimate it costs approximately $50 billion to establish one gigawatt of computing capacity, accounting for hardware, energy infrastructure, and data center development.​

Based on Sam Altman’s 250 GW target, the implied cost would reach a staggering $12.5 trillion!

Line chart on white background with gray gridlines displaying continuous growth from January 2024 to October 2025. Y-axis ranges from 0 to 800, X-axis shows monthly intervals. Blue line begins at approximately 100 in January 2024, rises gradually through 200 by mid-2024, reaches 300 by late 2024, then accelerates sharply through 2025, crossing 400 in January, 500 by April, 600 by July, and approaching 800 by October 2025
Exponential growth curve demonstrating 8x expansion from Q1 2024 baseline of ~100 to ~800 by October 2025. The trajectory shows steady linear growth through late 2024, followed by steeper acceleration beginning in early 2025, with the most aggressive growth phase occurring between April and October 2025—Reuters

There’s no assurance that OpenAI or other AI companies will secure the necessary funds to cover these expenditures.​

Citi estimates the company’s revenue will only rise to around $163 billion by 2030—a mere 13% of its expenses.​

This financing gap suggests that energy infrastructure investments may need to come from utilities, governments, and ultimately ratepayers rather than the AI companies themselves.

The artificial intelligence boom has created an unprecedented demand shock for electricity that most investors have failed to comprehend.​

Data centers consumed approximately 415 terawatt-hours globally in 2024, representing about 1.5% of total global electricity demand.​

By 2030, that figure is projected to more than double to 945 terawatt-hours—roughly equivalent to Japan’s entire electricity consumption.​

In the United States specifically, the growth trajectory is even more dramatic.

Goldman Sachs calls this,

“the most constrained infrastructure bottleneck in modern history.”

On November 9, 2025, Goldman Sachs Research published their updated data center power demand forecast: 175% surge by 2030 versus 2023 levels—the equivalent of adding electricity demand greater than the entire nation of France.​

Aianddatacenters
569KB ∙ PDF file
Download
Download

The U.S. grid currently operates at 62 gigawatts of data center capacity.​

By 2030, we need 170 gigawatts.​

That’s 108 gigawatts of new capacity. In five years!

​​Even the conservative case requires building capacity equal to South Korea’s entire grid. The aggressive case requires building Germany’s entire grid—just for AI.

During a labor shortage. With 10-15 year nuclear plant construction timelines.

Training sophisticated AI models demands immense computing power—GPT-4 training alone required around 30 megawatts of power, while OpenAI’s Stargate initiative anticipates multi-gigawatt data centers.​

A typical generative AI query through ChatGPT uses approximately 10 times the energy of a standard Google search, fundamentally changing the economics of digital services.​

By 2035, data centers are projected to account for 11.6% of all U.S. electricity demand, more than triple their 3.5% share today.​

Bar chart comparing US data center demand versus supply from 2023-2028, showing demand growing from 24 to 77.1 gigawatts while supply lags from 14.2 to 66.7 gigawatts, with persistent 9-11 GW shortfall indicated by purple arrows
US data center capacity shortfall projected to widen from 9.8 GW in 2023 to 10.4 GW by 2028, as demand growth (24 GW to 77.1 GW) significantly outpaces supply expansion (14.2 GW to 66.7 GW).

This represents a fundamental shift in the American energy landscape, creating opportunities in power generation, transmission infrastructure, and energy efficiency technologies that dwarf the potential returns from semiconductor investments alone.

The Grid Can’t Handle What’s Coming

The electrical grid infrastructure in the United States was not designed for the explosive growth in data center demand that AI is driving.​

Power shortages are already delaying data center projects by two to six years, creating a bottleneck that threatens the entire AI revolution.​

This year alone, electricity demand from data centers in the U.S. is estimated to rise 22%.​

The challenge isn’t just about generating more electricity—it’s about delivering that power where it’s needed, when it’s needed, with the reliability that AI operations demand.​

Large AI data centers can require dramatically more energy per square foot than traditional data centers.​

A five-acre data center augmenting central processing units with specialized graphics processing units might see its energy usage increase from 5 to 50 megawatts—a tenfold increase.​

The hyperscalers—the leading AI infrastructure developers scaling data center networks globally—are constructing facilities that dwarf anything previously built.​

Each of the top three hyperscalers’ largest U.S. data centers currently draw less than 500 megawatts of power, but the largest data centers they are constructing or planning to build require between 1 and 2 gigawatts.​

Some planned campus developments could consume 5 gigawatts or more, equivalent to the power needs of several medium-sized cities.​

“The future of AI will not be forged in code and large language models alone. It will be built with concrete, steel and silicone.”

—Dan Dee​

Stacked bar chart showing quarterly data center capacity from Q1 2023 to Q4 2026 segmented into three categories: Hyperscale/Cloud (dark blue, 14% CAGR), Non-Hyperscale/Cloud (light blue, 4% CAGR), and AI (purple, 98% CAGR), with Goldman Sachs estimates shown from Q1 2025 onward
Data center capacity by workload type from Q1 2023 to Q4 2026, showing hyperscale/cloud maintaining dominance at 14% CAGR while AI workloads explode at 98% CAGR, growing from ~1 GW to ~16 GW—Goldman Sachs Research

​This massive supply-demand imbalance creates an investment opportunity that transcends traditional utility sector dynamics.

The top five hyperscalers—Amazon , Microsoft, Google, Meta, and Oracle —spent $412 billion in combined capital expenditures in 2025 alone, with 60-70% earmarked for data center and power infrastructure.​

That’s $442 billion flowing into power infrastructure in a single year—more than the entire U.S. renewable energy sector invested over the past decade.​

Translation: AI revolution has moved from a software problem to a physics problem.​

Today, the hyperscalers have quietly flipped the switch. They’re no longer in “cost-optimization” mode when it comes to power. They’re in full-blown survival mode.

Microsoft, Google, Amazon, and Meta are now racing to lock in guaranteed, always-on electricity at almost any price — because the first ones to secure gigawatts of reliable baseload power will enjoy an unassailable, decade-long moat over everyone else.

In the AI arms race, megawatts have become the new oil.

The companies that own (or control) them don’t just win — they render latecomers mathematically incapable of catching up.

Power is becoming the ultimate barrier to entry!

Grouped bar chart of yearly capital spending from 2020 to 2025 for Amazon, Microsoft, Alphabet, Meta and Oracle, with 2025 in blue: Amazon rises from $40 billion to $132 billion, Alphabet from $22 billion to $91 billion, Microsoft from $18 billion to $83 billion, Meta from $15 billion to $70 billion, Oracle from $2 billion to $35 billion, and the total from $97 billion to $412 billion.
Big Tech's capex explosion: capital spending by Amazon, Microsoft, Alphabet, Meta and Oracle rose from $97 billion in 2020 to $239 billion in 2024 and $412 billion in 2025 (calendar years; blue = 2025). In 2025 Amazon spent $132 billion, Alphabet $91 billion, Microsoft $83 billion, Meta $70 billion and Oracle $35 billion. Source: company 10-K and 10-Q filings via SEC EDGAR XBRL (purchases of property and equipment; https://www.sec.gov/edgar/search/).

Let that sink in:

Tech companies are spending more on electricity infrastructure than semiconductors!

The deals prove desperation has replaced optimization:​

  • Amazon: $650 million for Talen Energy data center campus in Pennsylvania, paying 40% premium to market rates specifically for direct nuclear plant connection delivering 1,920 MW through 2042.​

  • Microsoft: $26.85 million for 1,350-acre North Carolina site near Duke Energy nuclear facilities, then negotiated separate 20-year power purchase agreement at undisclosed (likely premium) rates.​

  • Google: Signed multiple power deals totaling $3+ billion in commitments for data centers in Ohio, Indiana, and Virginia—all requiring new generation capacity.​

  • Meta: Advanced negotiations for multi-gigawatt agreements in Texas and Louisiana, reportedly offering take-or-pay provisions guaranteeing payment even if capacity isn’t fully utilized.​

When companies worth $8+ trillion combined are paying 40% premiums and offering guaranteed payment for unused electricity, that’s not normal corporate behavior.​

That’s existential panic.​

GS SUSTAIN: Data Center Power Demand-The 6 Ps Driving Growth and Constraints
1.56MB ∙ PDF file
Download
Download

Virginia’s Grid: Canary In The Data Center Coal Mine

The scenario is materializing in Northern Virginia—the world’s largest data center market with 13% of global operational capacity.​

A state-commissioned study by Energy + Environmental Economics (E3) just delivered a verdict that should terrify every tech CEO: Virginia’s power grid cannot support current data center commitments without unprecedented infrastructure investment!​

The report’s conclusion is devastating:​

“The sustained scale and pace of data center growth in the region will likely be constrained by new energy infrastructure development. Grid interconnection policies will delay new load from connecting until the infrastructure to serve it reliably is in place. These policies will constrain data center load growth to the rate of infrastructure expansion.

Building enough infrastructure for data center demand will be very difficult and meeting half that demand is still difficult”

Line chart titled 'Average monthly energy use (GWh)' for Virginia with Y-axis from 0 to 35,000 GWh and X-axis from 2023 to 2040. Four projection lines start at approximately 11,000 GWh in 2023. Top dark blue solid line labeled 'Unconstrained demand +183%' reaches 30,000+ GWh by 2040. Dark blue dashed line 'PJM forecast (adjusted) +152%' reaches approximately 27,000 GWh. Medium blue solid line 'Half of unconstrained demand +99%' reaches approximately 21,000 GWh. Light gray line 'No new data center demand +15%' shows minimal growth to approximately 12,000 GWh. Green triangle indicators mark percentage increases. Source: JLARC staff consultant analysis. JLARC logo (Joint Legislative Audit and Review Commission) in bottom right
Virginia energy demand projections through 2040 showing four scenarios: unconstrained data center demand (+183%), PJM forecast adjusted (+152%), half unconstrained demand (+99%), and no new data center demand baseline (+15%), sourced from JLARC analysis.

Translation: Data centers already approved and funded may never turn on because the electricity doesn’t exist.​​

And Northern Virginia represents just 13% of global data center capacity.​

What happens when this bottleneck hits every major market simultaneously?

The $5 Trillion Infrastructure Gap Nobody Can Fill

I’ve studied every major U.S. infrastructure cycle since 1900.

This AI power infrastructure wave has three characteristics that make it the most investable in modern history:

1. Demand Certainty Is Contractually Guaranteed

Past cycles: Infrastructure built hoping demand would justify it.​

  • Telecom companies laid 100 million miles of fiber in 1999-2001 betting internet traffic would grow exponentially.​

  • When dot-com bust hit, 70% of that fiber went unused for a decade.​

This cycle: Amazon, Microsoft, Google, Meta sign 20-year offtake agreements BEFORE utilities break ground.​

  • Constellation doesn’t build capacity hoping Microsoft will buy.​

  • Microsoft already committed to buying—in writing, through 2042.​

2. Competition Barriers Are Insurmountable

Semiconductor fabs: $10-20 billion, 3-5 years to build.​

  • Result: TSMC , Samsung , Intel all building competing fabs.​

Nuclear plants: $10-20 billion, 10-15 years to build, require NRC approvals, environmental reviews, local opposition battles.​

  • Result: Maybe 5 companies globally capable of executing.​ And hyperscalers need power in 2026-2028, not 2038.​

Existing capacity is a non-replicable monopoly.​

3. Multiple Expansion Hasn’t Happened

When investors realized software would dominate in 2002-2010, valuations re-rated from 12-15x earnings to 25-40x+.​

Some Energy infrastructure still trades at 10-20x despite growth profiles matching software’s peak.​

Once Wall Street reprices these from “boring utilities” to “AI infrastructure enablers,” we’ll see 50-100% multiple expansion on top of earnings growth.​

Constellation at 30x forward? That’s cheaper than Nvidia at 35x when you compare growth + certainty.​

Vistra at 17x with 22% projected EPS growth? Absurdly undervalued.​

The re-rating hasn’t happened yet.​

That’s the opportunity!

Goldman Sachs Research estimates $5 trillion in funding will be needed for digital infrastructure and power through 2030.​

McKinsey puts the broader energy transition requirement at $9.2 trillion annually through 2050—$3.5 trillion more per year than current spending.​

This is different from every infrastructure boom we have seen!

What makes the AI power buildout different from every previous infrastructure cycle?

The capital is already committed, not speculative.

Table with five columns titled Era, Project, Public, Private, and Total Spend (2024 adj.). Four rows of data: Row 1: 1950s-60s, Highways, 90%, 10%, approximately $600 B. Row 2: 1990s, Fiber Backbone, 60%, 40%, approximately $800 B. Row 3: 2010s, Renewables, 50%, 50%, approximately $1.2 T. Row 4: 2025-35, AI Power & Data Centers, 15%, 85%, approximately $5 T (est.). Table shows progression from public-dominated to private-dominated infrastructure funding over time
AI Infrastructure: $5 Trillion Private Capital vs. Historical Government-Led Buildouts.” Left side shows historical infrastructure waves with public/private funding mix: 1950s Highways (90% public, 10% private), 1990s Fiber (60% public via spectrum/subsidy, 40% private), 2010s Renewables (50% public via tax credits, 50% private). Right side shows 2025-2030 AI Power: 85% private ($412B hyperscaler capex + utility capex), 15% public (IRA credits, grid modernization). Bold text: “First major infrastructure wave in U.S. history driven by PRIVATE capital with revenue certainty BEFORE construction begins.

The 1990s fiber buildout?

Speculative—companies laid cable hoping internet traffic would materialize.​

Result: 2001-2002 telecom bust when demand didn’t justify investment.​

The 2025-2030 AI power buildout?

Pre-committed via 20-year offtake agreements signed BEFORE construction!

Amazon doesn’t pay $650 million Talen energy for data center hoping they’ll need power in 2030.​

They’re securing it because they already have AI deployment roadmaps requiring that exact capacity.​

This has NEVER happened in infrastructure history: Private companies committing hundreds of billions with multi-decade revenue certainty.​

Nuclear Renaissance Is Real (And Accelerating)

Line chart titled 'Terawatt-hours (TWh) of electricity demand, medium scenario' showing US data center energy consumption from 2023 to 2030. Y-axis ranges from 0 to 700 TWh. Black line with data points: 2023: 147 TWh (3.7% share); 2024: 178 TWh (4.3%); 2025: 224 TWh (5.2%); 2026: 292 TWh (6.5%); 2027: 371 TWh (8.0%); 2028: 450 TWh (9.3%); 2029: 513 TWh (10.3%); 2030: 606 TWh (11.7%). Below chart, table shows 'Share of total US power demand, %' for each year. Source: Global Energy Perspective 2023; McKinsey, October 18, 2023; McKinsey analysis
Data centers emerge as dominant US electricity consumer: McKinsey projects “medium scenario” growth from 147 TWh (2023) to 606 TWh (2030)—a 4.1x increase that will consume nearly 12% of total US power demand. The acceleration is remarkable: 147→178 TWh (+21%, 2023-24) compounds to 513→606 TWh (+18%, 2029-30), reflecting exponential AI infrastructure buildout. This 459 TWh incremental demand exceeds total annual consumption of most countries.

Nuclear energy is experiencing a remarkable resurgence as the only carbon-free baseload power source capable of meeting AI’s insatiable appetite for electricity.​

The U.S. government is aiming to quadruple nuclear energy capacity by 2050 and restart the domestic uranium industry to support soaring energy demand.​

Major technology companies have made substantial investments in nuclear power partnerships.​

Constellation Energy, the largest producer of carbon-free electricity in the United States, has emerged as a primary beneficiary of this nuclear renaissance.​

The company produces more than 32 gigawatts of capacity, with roughly 90% of its annual output being carbon-free, positioning it perfectly to meet the high-uptime, low-emissions requirements of AI data centers.​

Constellation’s stock has more than tripled since ChatGPT’s launch in November 2022.​

Line chart of Constellation Energy's share price in 2024, rising from $115 in January, jumping in late February and again on 20 September after the Three Mile Island deal with Microsoft, peaking at a record $286 on 4 October and ending the year at $224.
Constellation Energy shares in 2024: from $115 on 2 January to a record $286 on 4 October after the 20 September deal to restart Three Mile Island Unit 1 for Microsoft, ending the year at $224 (+94%). Source: Yahoo Finance (CEG).

The nuclear opportunity extends beyond established operators to next-generation technologies.​

Small modular reactors and microreactor technologies are attracting significant capital, with companies like Oklo (backed by Sam Altman) seeing their stock surge 500% in 2025 despite generating no revenue.​

The investment thesis is straightforward: nuclear provides the only proven technology capable of delivering gigawatt-scale, carbon-free, baseload power that AI infrastructure demands.​

The hyperscaler energy trilemma has only one solution:​

  1. Carbon-free (meets corporate climate pledges)

  2. Baseload (24/7 reliability AI training requires)

  3. Scalable (gigawatt-level capacity)

Solar and wind? Can’t provide 24/7 power.​

Natural gas? Carbon emissions violate sustainability commitments.​

Batteries? Can’t store gigawatt-scale for continuous AI training.​

Only nuclear checks all three boxes.​

Policy response has been unprecedented bipartisan acceleration:​

  • Biden administration: $900 million for nuclear development in Inflation Reduction Act​

  • Bipartisan Infrastructure Bill: $6 billion preventing nuclear plant closures​

  • Trump administration: Signaled support for Small Modular Reactor (SMR) development​

  • State legislatures: Virginia, Texas, Ohio fast-tracking nuclear permits for data center zones​

Even the Sierra Club—which spent 40 years fighting nuclear—now quietly supports existing plant life extensions because it’s the only path to AI growth AND decarbonization.​

Companies that own operational nuclear capacity—which takes 10-15 years and $10-20 billion to replicate —now control monopoly assets.​

There are maybe five companies globally capable of building new nuclear at scale.​

And hyperscalers need gigawatts today, not in 2040.​

Existing nuclear operators can name their price.​

Nuclear Stocks That Already Captured The Theme

While diversified investors debated “concentration risk”, nuclear energy stocks delivered the exceptional performance—fueled by indisputable AI boom. ​

  • Constellation Energy : Up 62% YTD through October 2025, missed Q3 earnings but stock reversed higher same-day as investors prioritize 20-year hyperscaler power contracts over quarterly results—market betting on multi-decade contracted cash flows from AI infrastructure buildout.​

  • Vistra : Up 85% in 2025 (top 10 S&P 500 performer) despite revenue miss, just announced 1,200 MW, 20-year contract for Comanche Peak nuclear plant at undisclosed premium pricing—validating thesis that nuclear power commands premium from hyperscalers seeking carbon-free baseload capacity.​

  • Talen Energy : Up 75% in 2025, beat analyst expectations despite Amazon direct-supply deal facing regulatory challenges—proving direct-to-hyperscaler business model viable even with initial regulatory headwinds and establishing blueprint for behind-the-meter data center arrangements.​

  • GE Vernova : Up 380% since April 2024 spinoff, trading at 30x forward earnings with $36-37B revenue guidance and multi-year order backlog—premium valuation justified by grid infrastructure and power generation equipment demand accelerating from AI buildout and broader electrification trends

Line chart from January 2024 to 18 November 2025, rebased to 100: an AI power basket of Constellation, Vistra and Talen, in blue, climbs to about 520 before ending at 446; a semiconductor basket of Nvidia, Broadcom and AMD ends at 285; the S&P 500 ends at 142.
Power beats chips in the AI build-out: from the start of 2024 to 18 November 2025, an equal-weight AI power basket (Constellation Energy, Vistra, Talen Energy) returned +346%, a semiconductor basket (Nvidia, Broadcom, AMD) +185% and the S&P 500 +42%, with dividends reinvested. Source: Yahoo Finance (adjusted closes; S&P 500 Total Return Index, ^SP500TR).

Analysts in November 2025 captured the dynamic perfectly:​

“The market is applying a generous discount rate to the sector’s near-term risks, prioritizing its future promise. Strong share price appreciation and significant strategic deals coexist with mixed short-term financial results. This divergence suggests investors must weigh powerful policy tailwinds against erratic quarterly performance.”

Translation: The market has moved from pricing quarterly earnings to pricing 20-year contracted revenue streams.​

That’s not speculation—it’s re-rating an entire sector from “boring utility” to “AI infrastructure enabler”.​

Why Energy beats Semiconductors in 2024/2025?

Simple, in five words: A 20-Year Contract Moat.

Yes, Nvidia has been spectacular—up 3,000%+ over five years.​

But it trades at 40-50x forward earnings with zero revenue visibility beyond 12-18 months.​

Energy infrastructure plays trade at 10-32x with contractually guaranteed revenue through 2042-2045.​

Energy Infrastructure vs. Semiconductors — factor comparison chart for energy infrastructure vs semiconductors.
Energy Infrastructure vs. Semiconductors

​This is the “quiet” part, I think, the market hasn’t fully priced yet:

The companies that control the electrons have signed decades long, inflation-protected, zero-switching-cost monopolies… and the market is still giving them utility multiples while handing 50× earnings to companies that can be disrupted in a single node cycle.

That’s why the green line on the chart went outperforms — and why the divergence still has years left to run.

When Amazon commits to $650 million for 1,920 MW through 2042, that’s not a purchase order that can be canceled if AI spending slows.​

It’s a 20-year revenue annuity backed by an $2.5 trillion market cap company.​

When Vistra signs 1,200 MW for Comanche Peak through 2045, they’re not hoping demand materializes.​

They’ve already locked in the cash flows.​

Nvidia sells chips that become obsolete in 24-36 months and face AMD competition, custom silicon threats, and geopolitical China risk.​

Constellation owns nuclear plants that cannot be replicated in less than 10-15 years and face near-zero competition.​

One is a product cycle.

The other is a monopoly asset.


Join 1,000+ Investors Using the 80/20 Rule to Build Smarter, Higher-Returning Portfolios

The Pareto Investor delivers a proven, systematic framework to apply the 80/20 rule to stocks — focusing only on the tiny fraction of companies that drive nearly all market gains.

Membership Tiers for 2026:

Premium — Everything you need to outperform the market with less effort:

  • Full access to all live Pareto portfolios (Growth, Dividend & Permanent)

  • Monthly portfolio updates with exact stock selections

  • Rebalancing alerts & detailed review framework

  • Exclusive investment insights & advanced portfolio optimizations

Elite — Everything in Premium & permanent ownership of my most powerful tools:

  • Pareto Elite Sheet — Fully automated annual rebalancing, real-time portfolio dashboard, and live tracking.

  • 50+ Years of Historical Stock Data.

  • Pareto Watchlist & AI Stock Screener.

  • EXCLUSIVE All-Weather Portfolio — Immediate, exclusive access to my stock-based portfolio built to hold up in every market. Available only to Elite members.

Only 4% of stocks account for nearly all net wealth creation in the market. Why settle for mediocre index returns held back by thousands of underperformers? Join 1,000+ investors focusing only on the rare winners — and upgrade your returns with the Pareto Principle today.

Subscribers’ reviews—2026


Get 50% off forever


I’ll Be Direct About What’s Happening

Your electricity bill is about to spike 30-50% over the next five years.​

And energy investors will profit from every dollar of that increase.​

Line chart of the average U.S. city electricity price from 2014 to September 2026, flat between about 13 and 14 cents per kWh until 2020, then climbing steadily from 13.6 cents in January 2021 to 19.6 cents.
U.S. electricity prices were flat from 2014 to 2020, between $0.131 and $0.143 per kWh, then climbed 44% from $0.136 in January 2021 to $0.196 in September 2026. Source: U.S. Bureau of Labor Statistics, average price data, via FRED (APU000072610; https://fred.stlouisfed.org/series/APU000072610).

Here’s the regulatory model that creates guaranteed wealth:​

  1. Utility invests billions building generation capacity for data center customers

  2. Earns guaranteed 9-11% regulated return on that investment

  3. Bills ALL ratepayers—residential, commercial, industrial—not just data centers

  4. Repeat annually for decades

Every dollar gets recovered from ratepayers plus the regulated return.​

Energy stocks don’t just win if AI succeeds.​ They win regardless—because the infrastructure gets built and ratepayers fund it.

Building A Strong Energy Portfolio For AI Boom

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2026 The Pareto Investor · Publisher Privacy ∙ Publisher Terms
Substack · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for writing, podcasting, and media founders.