Hey there, bargain hunter.

Here is a question worth sitting with: What happens when a system that was designed for one era gets asked to do something it was never built for?

That is exactly what is playing out right now across the US power grid. Not slowly. Not theoretically. Right now, this summer, in real time.

Back-to-back heat waves are pushing electricity demand to levels that are breaking records set two decades ago. On July 2, PJM Interconnection, the country’s largest grid operator serving more than 65 million people across 13 states and Washington D.C., saw usage soar to an unprecedented 168 gigawatts, topping a high that had stood for twenty years. Temperatures across the eastern seaboard hit triple digits. Air conditioners ran at full capacity. And sitting quietly underneath all of that residential demand was a completely different and rapidly growing load: AI data centers requiring around-the-clock power regardless of what the thermometer reads.

The collision was ugly enough that the Department of Energy stepped in directly. Energy Secretary Chris Wright directed data centers in the mid-Atlantic region to switch to backup diesel generators to free up grid capacity for residential air conditioning. That is not a normal market event. That is a stress signal.

What the Numbers Actually Say

Let’s get specific, because the scale here is genuinely hard to absorb.

US grid power supplied to data centers increased 25% in 2025, reaching roughly 64.4 gigawatts, and has nearly tripled since 2020. Goldman Sachs projects that figure more than doubles again to 66 GW by 2027. S&P Global’s 451 Research puts data center grid demand at 75.8 GW in 2026 and climbing toward 134 GW by 2030. BloombergNEF goes further, estimating demand from American data centers could reach 194 gigawatts by 2035.

That last number represents roughly the equivalent of 194 nuclear power plants running continuously.

The IEA found that data centers drove roughly half of all US electricity demand growth in 2025, and expects that share to hold through 2030. GPU-accelerated servers handling AI workloads now account for around 60% of electricity consumption in modern data centers, with cooling systems adding another 7% to 30% on top of that depending on the facility.

Meanwhile, the grid trying to absorb all of this was largely designed and built during a period when electricity demand was flat for two decades.

The PJM capacity market is where you can see the price of that mismatch most clearly. The 2026/2027 capacity auction cleared at a record $329.17 per MW-day, a 22% increase from the prior year. The year before that, prices had already jumped nearly ninefold. Total annual customer costs in PJM have risen from roughly $2.2 billion in 2024 to $14.7 billion, and the most recent auction for 2027/2028 pushed costs to $16.4 billion. The auction cleared with just 139 MW of additional capacity above the reliability requirement. That is not a comfortable margin for a system serving tens of millions of households.

Slight tangent, but it matters: ERCOT in Texas is projecting demand could surpass 92 gigawatts this summer, above the state’s current record of 85.5 GW set during the August 2023 heat wave. The EIA projects Texas grid demand rises 7% in 2025 and another 14% in 2026 as data centers and other large loads come online. The stress is not confined to one region.

Where the Market May Be Getting It Wrong

Here is where The Cheap Investor’s job actually begins.

The obvious trade on power grid stress is the high-flying infrastructure contractors and AI-adjacent names that have already been discovered. Quanta Services (PWR) has a forward P/E sitting near 55, well above its five-year average of roughly 24.5. The business is excellent. The tailwinds are real. But at that valuation, a lot of good news is already priced in, and the margin of safety is thin for a patient value investor.

The less-discussed opportunity may sit in the regulated utilities directly absorbing the load growth and spending the capital to handle it. These are not flashy names. They are not momentum stocks. But the fundamental shift in their earnings trajectory over the next five years is significant, and in at least one case, the market appears to be only partially pricing that in.

Duke Energy: The Toll Bridge Nobody Is Talking About

Duke Energy (DUK) serves approximately 8.7 million customers across North Carolina, South Carolina, Florida, Indiana, Ohio, and Kentucky. That footprint sits squarely in the Southeast AI corridor, the region where hyperscalers are racing to place new data centers due to favorable regulatory conditions, land availability, and relatively competitive power costs.

The Q1 2026 numbers came in ahead of expectations. Adjusted earnings per share of $1.93 beat the consensus forecast of $1.86 by nearly 4%, while revenue of $9.18 billion surpassed estimates by more than 8%. Full-year 2026 EPS guidance of $6.55 to $6.80 was reaffirmed, and the company reported 2025 EPS of $6.31, up 7% year over year.

The growth behind those numbers is not a one-quarter fluke. Duke signed 4.5 gigawatts of electric service agreements with data centers, and management flagged another 7.8 GW in its high-confidence late-stage pipeline. The company raised its five-year capital plan by $16 billion to $103 billion, targeting grid upgrades, generation buildout, and fuel infrastructure investment. That plan is projected to drive roughly 9.6% earnings base growth over the period, with approximately 14 GW of incremental generation added over five years.

Management highlighted that electricity demand tied to AI data centers is growing at roughly ten times historical rates. They are not exaggerating.

The valuation picture is where it gets interesting for a value-oriented reader. Duke’s trailing P/E sits near 18.5, below the broad market multiple by a wide margin, and the stock’s EV/EBITDA was approximately 7.56 as of mid-May 2026. The company has paid dividends for 100 consecutive years. Current yield runs approximately 3.4% to 3.5% with a payout ratio around 65%, and the five-year EPS growth commitment is 5% to 7%, with management expressing confidence in the top half of that range beginning in 2028.

Analyst consensus price targets hover around $134 to $138, versus a recent trading range in the mid-$120s. That implied upside is not dramatic, but the combination of a growing regulated rate base, locked-in data center contracts with minimum billing provisions, and a credible multi-year capital deployment plan makes the earnings visibility here unusually strong for a utility.

The Cheap Investor Scorecard

• Business Quality: Regulated monopoly with irreplaceable regional infrastructure. High.
• Financial Strength: $103B capex plan funded by regulated rate base and contracted demand. Revenue up 11% YoY in Q1 2026. Solid.
• Valuation: Forward P/E near 18.5, EV/EBITDA near 7.5x. Reasonable relative to growth profile and 100-year dividend history.
• Competitive Position: Geographically dominant in Southeast data center corridor. Hard to replicate.
• Cash Flow: Regulated returns limit near-term free cash flow, but minimum billing provisions on data center contracts reduce revenue risk.
• Management Execution: Consecutive earnings beats, capital plan raised with specificity. Constructive.
• Catalyst Strength: Data center load ramp from 4.5 GW of signed ESAs already underway. Near-term and visible.
• Margin of Safety: Modest at current prices. Not cheap in absolute terms, but defensible relative to growth.
• Long-Term Potential: 5-7% EPS growth, potentially accelerating to the high end as data center loads ramp post-2027. Strong.

What Could Go Wrong

Regulatory risk is central to any regulated utility thesis. State commissions in North Carolina, South Carolina, and Florida control the allowed return on Duke’s rate base. A hostile rate case outcome could compress earnings and delay capital recovery. That risk is real and should not be dismissed.

Capital intensity is the other concern. A $103 billion plan over five years requires consistent access to debt markets. Higher interest rates directly raise Duke’s financing costs and can pressure earnings if rate case outcomes lag capital deployment timing.

Finally, the Goldman Sachs note that only about 50% to 60% of data center capacity scheduled for the next one to two years is expected to come online on schedule is a legitimate caution. If the AI buildout slows or contracts get delayed, the load ramp Duke is counting on takes longer to materialize. That would not break the thesis, but it would push the timeline out.

Bull, Base, and Bear

• Bull: Data center load ramps faster than expected. Rate cases come in constructive. EPS growth hits the top of the 5-7% range through 2030. Stock re-rates toward 20-21x earnings, implying meaningful upside from current levels.
• Base: EPS compounds at 6% annually through 2030. Dividend grows in line. Total return of 9-10% annually including yield. Modest but durable.
• Bear: Adverse regulatory ruling delays cost recovery. AI buildout decelerates. EPS growth stalls near the low end of guidance. Stock stays range-bound. Dividend remains intact, but capital appreciation is limited.

The broader story here is not complicated. The US power system was built during a long period of flat demand and is now being asked to absorb a structural demand surge that utilities, grid operators, and regulators were not fully prepared for. Utilities across the country have already warned that electricity demand is growing for the first time in decades, driven by AI data centers, electrification of transportation, and new manufacturing. That shift forces utilities to delay the retirement of aging plants while accelerating investment in new generation and transmission.

That investment cycle is real, it is funded, and it is long. The question for investors is not whether it happens. It is who gets paid to execute it, at what valuation, and with what margin of safety.

Duke Energy is not the cheapest stock in the market. But in a world where reliable compounding at a fair price is genuinely hard to find, a regulated monopoly sitting at the center of the AI power buildout, trading at a below-market multiple with locked-in demand growth and a century of uninterrupted dividends, is worth a serious look. Not a rushed one. A serious one.

The grid needed an upgrade long before anyone trained a language model. The bill is just finally coming due.

Stay patient,
The Cheap Investor