Cover Image

The Binding Constraint: What Power Shortages Mean for the AI Boom

November 27, 20255 min read

The week ending November 28, 2025 offered a striking demonstration of market volatility. After falling last week on renewed concerns about an AI bubble, equities rebounded sharply. The S&P 500 recovered its 2 percent decline and the NASDAQ erased nearly 3 percent of losses. Comments from several Federal Reserve governors hinting at a potential December rate cut helped fuel the advance. Despite the turbulence, markets remain close to their all time highs.

Earnings reports were limited during the Thanksgiving holiday period, but one name drove headlines.


Dell’s Strong Guidance Signals AI Tailwinds

Dell reported fiscal third quarter results that were mixed on the surface. Revenue grew 11 percent year over year but slightly missed estimates. Earnings per share rose 17 percent and came in modestly ahead of expectations.

The surprise came in forward guidance. Dell expects approximately 31.5 billion dollars in fourth quarter revenue, far above the 27.6 billion analysts anticipated. It also raised its EPS outlook.

Dell benefits from rising demand for servers driven by AI infrastructure. The constraint is strategic rather than operational. The company has few proprietary products and must compete primarily on price and service. As a result, despite mid to high teens earnings growth, Dell trades at around 11 times forward earnings. Super Micro trades at similar levels. These multiples resemble banks, not high growth technology firms.


A Second Railroad in the AI Race

Last week I drew a historical parallel between today’s AI buildout and the railroad boom of the late nineteenth century. Railroads transformed the United States but did not guarantee profitability. Overcapacity, financial distress, and bankruptcies were common.

This analogy became more relevant after a memo circulated from Sam Altman to OpenAI employees. Google’s release of Gemini 3 appears to have surpassed ChatGPT in several important dimensions. Altman acknowledged the competitive pressure directly.

For years the narrative centered on nimble upstarts disrupting slow incumbents. That logic does not apply here. Google, Microsoft, and Amazon are not passive defenders of legacy markets. They responded quickly and decisively.

The financial comparison underscores the imbalance. In the third quarter of 2025, Google generated 102 billion dollars in revenue and 35 billion dollars in net income. It will spend more than 90 billion dollars on AI this year and can fund that internally. Cash remains near 100 billion and long term debt has barely increased.

OpenAI expects roughly 13 billion dollars in revenue for 2025 and plans to spend hundreds of billions in pursuit of leadership. That ambition depends entirely on capital raising. The second railroad has effectively caught up.


The Real Bottleneck: Power, Not Chips

Michael Burry recently criticized hyperscalers for extending depreciation schedules to boost reported earnings. It is a valid concern. Asset impairments could occur. BYD’s recent write down in China is an example.

However, the more fundamental issue is whether AI investments will produce the transformative gains many expect. If they do, the accounting debate will be secondary. If they do not, none of the spending will be justified. At this stage the outcome remains uncertain.

One constraint, however, is already visible.

The limiting factor is electricity.

A year ago the scarcity was semiconductor supply. That is no longer true. Nvidia, AMD, and other chipmakers are delivering at scale. The new bottleneck is the power required to operate data centers.

The United States will generate roughly 4,200 terawatt hours of electricity in 2025. Before 2021 demand grew about 1 percent annually. Now it grows about 3 percent. That 3 percent represents roughly 125 terawatt hours of incremental demand per year, the equivalent of powering several major cities. New York uses about 60 terawatt hours annually.

Data centers currently account for an estimated 5 to 7 percent of US electricity use. By 2028 that share is expected to exceed 10 percent.

The strain is already evident. In Santa Clara, two newly constructed 48 megawatt data centers remain idle due to the absence of available power. The local utility’s required grid upgrades will not be completed until 2028. Texas, another favored destination for hyperscalers, faces similar challenges. If all proposed projects were built, they would consume the equivalent power of 150 million homes. The Texas grid currently serves 30 million.

This is why hyperscalers are exploring unconventional solutions, including nuclear microreactors and vertically integrated power systems. Companies such as Oklo have surged in market value despite having no approved designs or operating revenue. The demand is clear. The certainty of supply is not.

Without power, AI infrastructure sits unused. Power is the defining constraint and expanding capacity requires years of planning, permitting, and construction.


Leadership Fatigue and the Limits of CEO Mythology

The United States often elevates CEOs to celebrity status. Yet even highly regarded leaders can leave behind structural issues that surface later. Jack Welch’s celebrated tenure at GE ended with hidden liabilities that contributed to the company’s long decline. Brian Niccol’s departure from Chipotle was followed by significant share price weakness. Starbucks, where he now serves as CEO, continues to confront deep structural challenges.

Leadership effectiveness has a shelf life. Market narratives often overlook that reality.


Why Europe Continues to Lag

A viewer recently shared an experience that reflects structural challenges within European economies. After submitting a detailed business plan and sufficient capital to launch a modest consumer product venture, he saw his application rejected twice. He is now in court, losing valuable time and momentum.

His conclusion was straightforward. Europe has resources but a regulatory mindset that discourages entrepreneurial risk. In the United States such a proposal would face oversight but rarely outright rejection. This distinction helps explain Europe’s prolonged stagnation relative to US economic performance.


Final thoughts

The momentum behind artificial intelligence is tremendous. Capital is flowing. Chips are available. Infrastructure is being built at unprecedented speed. But the industry’s progress now runs into a constraint far more fundamental than hardware availability. Without sufficient and reliable power, data centers cannot operate. The future of AI depends on the speed at which the world can expand its electricity infrastructure.

Power, not chips, is the true binding constraint.

If you want more of these weekly wrap-ups, interviews and financial literacy content, check out our YouTube channel and realismanplaybook.com. And I’ll see you next week.


Thanks for reading this week’s wrap.
If you’d like to catch my interviews and market breakdowns, visit The Real Eisman Playbook or subscribe to the Weekly Wrap channel on YouTube.


This post is for informational purposes only and does not constitute investment advice. Please consult a licensed financial adviser before making investment decisions.

I’m Steve Eisman, an investor and fund manager best known for predicting the 2008 housing market collapse. I’ve spent my career studying markets, risk, and the psychology that drives financial decisions. Today, I continue to invest and share lessons from decades of watching cycles repeat.

Steve Eisman

I’m Steve Eisman, an investor and fund manager best known for predicting the 2008 housing market collapse. I’ve spent my career studying markets, risk, and the psychology that drives financial decisions. Today, I continue to invest and share lessons from decades of watching cycles repeat.

LinkedIn logo icon
Instagram logo icon
Youtube logo icon
Back to Blog