You're Not in the AI Business. You're in the Real Estate Business.

There's a scene in The Founder that I think about a lot.

Ray Kroc, played by Michael Keaton, is struggling to make McDonald's profitable. He's got the franchise model working — restaurants are opening, burgers are selling — but the money just isn't there. He's still essentially broke. Then Harry Sonneborn, a former Vice President of Finance at Tastee-Freez, sits him down and delivers what might be the most important business insight of the 20th century:

"You're not in the burger business. You're in the real estate business."

Sonneborn's actual quote was even more pointed: "We are not technically in the food business. We are in the real estate business. The only reason we sell fifteen-cent hamburgers is because they are the greatest producer of revenue, from which our tenants can pay us our rent."

The idea was simple but radical. Instead of just licensing the McDonald's brand to franchisees, the corporation would buy or lease the land underneath every restaurant. Franchisees would then pay rent to McDonald's on top of their franchise fees. The burgers were just the mechanism that generated the rent checks.

Today, McDonald's owns 80% of the buildings and 56% of the land across over 43,000 locations worldwide. Their real estate portfolio is valued at approximately $120 billion. In 2024 alone, McDonald's made over $7 billion from real estate — and they kept about 80 cents of every dollar from franchisees as operating profit, compared to just 15 cents from company-operated stores. The burgers are almost a rounding error.

I think something very similar is happening right now with artificial intelligence.

The AI Land Grab

Everyone's talking about AI models. Which one is smarter, which one is faster, which one can write better code or generate better images. OpenAI vs. Anthropic vs. Google vs. Meta vs. DeepSeek. It's the AI version of arguing about whose burger tastes better.

But while we're all debating model benchmarks, the real money is quietly flowing somewhere else entirely: into land, concrete, power plants, and cooling systems.

The numbers are frankly hard to comprehend.

Microsoft announced plans to invest $80 billion in AI infrastructure CapEx in 2025. Meta committed $60 to $65 billion. Google and Amazon are in the same ballpark. The Stargate Project, a joint venture between OpenAI, SoftBank, Oracle, and MGX, is targeting $500 billion in AI infrastructure investment. Brookfield Asset Management launched a $100 billion AI infrastructure program backed by NVIDIA and the Kuwait Investment Authority. Data center mergers and acquisitions hit a record $57 billion in 2024, with private equity driving the majority of deals.

Add it all up and analysts like BCG project the AI industry will spend roughly $1.8 trillion on infrastructure by 2030. For context, McDonald's $120 billion real estate empire took 70 years to build. The AI industry is building something an order of magnitude larger in roughly six.

These aren't software investments. This is land acquisition, construction, and energy procurement. This is real estate.

Follow the Stock Prices

If you want to know where the real value is accumulating, don't look at the AI companies. Look at the infrastructure companies.

Constellation Energy — a nuclear power company — has seen its stock surge roughly 300% over the past three years, peaking above 450% in late 2025. It was one of the top-performing stocks in the entire S&P 500 in 2024, outpacing Meta at 277% and Microsoft at 61% over the same three-year period. Read that again. A nuclear utility company is delivering returns that rival the biggest names in tech. Even NVIDIA, the undisputed chip king at roughly 720% over three years, reinforces the infrastructure thesis — its GPUs are worthless without the data centers to house them.

Harry Sonneborn would understand exactly why.

Data center REITs like Equinix and Digital Realty have been on a tear. Blackstone and CPP Investments paid roughly $16 billion to acquire AirTrunk, which operates hyperscale data centers across the Asia-Pacific region. Brookfield already has over $100 billion invested in digital infrastructure globally and is raising more.

And then there's the land itself. A 124-acre parcel in Prince William County, Virginia — part of the "Data Center Alley" corridor — sold to Microsoft for $3.75 million per acre. Ranchland in Texas has seen 10x value increases when earmarked for data center development. Farmers and ranchers who couldn't give away their land five years ago are now sitting on goldmines — not because of what's under the soil, but because of what can be built on top of it.

This is the McDonald's playbook, just at a completely different scale.

The Power Problem Is Really a Real Estate Problem

Here's where it gets really interesting. The biggest constraint on AI growth right now isn't model architecture or training data or even GPU supply. It's power.

Goldman Sachs estimates that global data center power usage currently sits at about 55 gigawatts, with AI representing roughly 14% of that. By 2027, they project total usage will reach 84 gigawatts, with AI growing to 27%. That kind of power demand doesn't come from software improvements. It comes from physical infrastructure — power plants, transmission lines, substations — all of which need to be built on land that needs to be acquired and permitted.

The tech giants know this, which is why they've collectively become some of the largest energy buyers on the planet:

Microsoft signed a 20-year deal with Constellation Energy to restart Three Mile Island Unit 1, securing 837 megawatts of carbon-free power by 2028. Yes, that Three Mile Island.
Google signed the world's first corporate agreement to purchase nuclear energy from small modular reactors built by Kairos Power — 500 megawatts across 6-7 reactors, with the first coming online by 2030 and full deployment by 2035.
Amazon agreed to buy 1.9 gigawatts from Talen Energy's nuclear plant in Pennsylvania and invested $500 million in X-energy to bring over 5 gigawatts online by 2039.
Meta inked a 20-year deal with Constellation Energy for the Clinton Clean Energy Center.

In total, big tech companies signed contracts for more than 10 gigawatts of new U.S. nuclear capacity in the past year. These are not software contracts. These are decades-long commitments to physical infrastructure on physical land.

When Microsoft is restarting a nuclear power plant to run AI models, you're not in the software business anymore. You're in the energy business. And the energy business is, fundamentally, a real estate business.

The Gold Rush Analogy (But Make It Real Estate)

The "picks and shovels" analogy gets thrown around a lot in tech. During the California Gold Rush, the people who made the real money weren't the miners — they were the ones selling picks, shovels, jeans, and provisions. Levi Strauss arrived to sell dry goods and died with a fortune worth hundreds of millions in today's dollars. He never mined a single ounce of gold.

The AI version of this analogy usually points to NVIDIA as the picks-and-shovels winner. And that's true — NVIDIA has done incredibly well selling the GPUs that power AI training and inference. But I think the analogy actually goes one level deeper than that.

In the Gold Rush, the picks-and-shovels sellers did well. But you know who did even better? The people who owned the land where the towns were built. The people who owned the saloons, the general stores, the boarding houses. The landlords. San Francisco's real estate boom wasn't a side effect of the Gold Rush — it was the enduring legacy of it. Long after the gold ran out, the land retained its value.

The same dynamic is playing out now. NVIDIA's GPUs will eventually face competition — from AMD, from custom silicon at Google and Amazon, from whatever comes next. But the land that data centers sit on? The power purchase agreements? The water rights for cooling systems? Those are much harder to replicate and much more durable.

Brookfield isn't investing $100 billion because they think one particular AI model is going to win. They're investing because regardless of which model wins, it's going to need a building to run in, land to sit on, and power to keep the lights on.

DeepSeek Proved the Point

When DeepSeek released its R1 model in January 2025, claiming performance comparable to OpenAI's o1 at a fraction of the cost — trained for roughly $6 million versus hundreds of millions — it wiped over half a trillion dollars from NVIDIA's market cap in a single day. The fear was that if AI models could be trained more efficiently, the demand for expensive GPUs and massive data centers would crater.

The market recovered quickly, and here's why: more efficient models don't reduce infrastructure demand. They increase it.

This is Jevons Paradox in action. When something becomes cheaper and more efficient, people use more of it, not less. DeepSeek didn't make data centers obsolete — it made AI accessible to a much larger market of potential users, all of whom need infrastructure to run their models on. Meta and Microsoft both confirmed that DeepSeek's emergence didn't change their infrastructure spending plans one bit.

Goldman Sachs projects data center occupancy rates will hit 95% by late 2026. The buildings are filling up faster than they can be built.

If you're a real estate investor, this is the kind of demand curve you dream about.

What McDonald's Can Teach the AI Industry

Let's bring this back to Harry Sonneborn. His insight wasn't just that real estate was profitable — it was that real estate was the durable competitive advantage. Burger recipes can be copied. Restaurant designs can be replicated. But if you own the land underneath the restaurant, you have leverage that no competitor can take away from you. The franchisee can't just leave — they'd lose their location, their customer base, everything.

The same logic applies to AI infrastructure. AI models are improving rapidly, costs are dropping, open-source alternatives are proliferating. The model layer is becoming increasingly commoditized. But the physical layer — the land, the power, the cooling, the fiber connectivity — is getting more scarce and more valuable by the day.

The companies that understand this are already positioning themselves:

NVIDIA isn't just selling chips — it's co-investing in Brookfield's $100 billion infrastructure fund
Microsoft isn't just building Copilot — it's restarting nuclear reactors
Amazon isn't just offering Bedrock — it's buying gigawatts of power capacity
Meta isn't just training Llama — it's building Prometheus, a gigawatt-scale data center campus in Ohio

These companies have already internalized the Sonneborn insight. The AI models are the hamburgers. The data centers are the real estate.

What This Means for the Rest of Us

I'm not writing this as an investment advisor (please, consult an actual one). But I do think this framing is useful for anyone trying to understand where the AI industry is actually heading.

If you're building AI products, understand that your competitive moat probably isn't your model. Models are getting cheaper and more accessible every month. Your moat is more likely to be your data, your distribution, and — increasingly — your access to compute infrastructure.

If you're thinking about the AI industry from an investment or career perspective, it's worth considering the less glamorous parts of the stack. The companies building power plants, laying fiber, constructing cooling systems, and acquiring land aren't the ones getting breathless TechCrunch coverage. But they might be the ones generating the most durable returns.

And if you're a landowner in Northern Virginia, Central Texas, or anywhere near a major power substation — you might be sitting on something more valuable than you realize.

McDonald's didn't win the fast food wars because they had the best burger. They won because Harry Sonneborn understood that the real business was the one nobody was paying attention to.

The AI industry hasn't fully learned this lesson yet. But the smart money already has.

Building AI-powered products and thinking about infrastructure? Find me on LinkedIn. Always happy to talk shop about the intersection of AI, infrastructure, and why the boring stuff is usually the important stuff.

Sources

McDonald's Real Estate:

AI Infrastructure Investment:

Nuclear Power Deals:

Data Center Power & Occupancy:

DeepSeek & Market Impact:

Stock Performance:

Meta Prometheus:

Meet Prometheus: World's Highest Capacity Data Center (NBC4)

Gold Rush Parallel: