#49 | AI's $100B shell game? You decide.

TL;DR: The biggest AI deal (Nvidia, OpenAI) in history looks like competition, but it’s actually the same money moving in very sophisticated circles.

👋 Hello,

We probably all know a trendy restaurant in our city that everyone raves about.

Perfect reviews everywhere. The delivery app keeps suggesting it. Food bloggers can’t stop talking about their “authentic flavors” and “hidden gem” status. Even that local foodie newsletter you get offers discounts there.

Then you discover something that makes you do a double-take.

Weirdly, the same owner controls the restaurant, the delivery platform, the review site, the food blog, the food supplier, the newsletter, and maybe even the “independent” local magazine that featured it.

Suddenly, what felt like organic buzz reveals itself as… well, something else entirely.

Ok, the food might still be good. And the reviews might reflect real experiences. But the whole ecosystem that made you discover and trust this place? That was choreographed.

Last week, when reading about AI’s biggest deal in history, I got similar vibes or an uncomfortable ping of recognition.

The AI Learning Guy newsletter 🤖 🧠💡

AI learning hacks and mega prompts delivered to your inbox.

Following the money in loops

NVIDIA just announced a $100 billion commitment to OpenAI. Many headlines painted it as proof of massive market demand, validation of AI’s future.

Ok, sounds great. But here’s what could also be true.

NVIDIA writes OpenAI a check for $10 billion initially, with up to $100 billion more as infrastructure gets deployed.

OpenAI takes that money and leases NVIDIA’s data center capacity and chip infrastructure.

NVIDIA’s stock jumps $170 billion on news of what looks like tremendous demand.

Now, here’s where it gets interesting: NVIDIA funded over 50 other AI companies in 2024. Those companies also spend their funding on… NVIDIA infrastructure.

See the pattern emerging? What appears to be market validation might largely be NVIDIA’s own capital flowing back to them through carefully structured loops.

Financial analysts have started using terms like “circular financing.” One called it “bubble-like behavior.”

Your restaurant discovery, but with semiconductors and hundred-billion-dollar stakes.

Standing in Abilene, consuming a nation’s electricity

Sam Altman chose an oddly humble backdrop for his announcement. Standing on dusty ground in Abilene, Texas—population 130,000—he outlined infrastructure that will consume as much electricity as 8 million homes.

The contrast struck me as almost comical. Endless Texas sky. Heat that could fry an egg on concrete. Behind him, the outline of what will become computational infrastructure more complex than most space programs.

First gigawatt comes online in the second half of 2026, powered by NVIDIA’s Vera Rubin platform. Named after the astronomer who discovered dark matter—fitting, maybe, for infrastructure most people will never see.

Think about what most of us experience when we use AI. Type a question into the chat box. Get an answer. Clean, simple, boring.

Behind that mundane interface: the electrical equivalent of medium-sized countries, millions of specialized chips, and financial arrangements that exceed many nations’ annual budgets.

Altman calls this “abundant intelligence.” The abundance seems to require resource scarcity, which makes nuclear programs look modest.

I could be wrong, but something about that math feels off.

The ecosystem that might validate itself

This could extend beyond one partnership announcement.

What we might be observing is the construction of an AI-industrial complex where companies serve simultaneously as customers, vendors, investors, and validators of each other’s success.

OpenAI committed $300 billion to Oracle over five years starting in 2027. Oracle invests in the Stargate project alongside OpenAI. SoftBank invested in OpenAI, participates in Stargate, funds infrastructure OpenAI will use.

Every dollar potentially creates its own demand signal. Every investment possibly generates its own success metrics.

Meanwhile, China’s DeepSeek achieves what appears to be comparable results using a fraction of computational resources. Their efficiency-first approach suggests that massive infrastructure might be more about financial engineering than technical necessity.

Though I’m probably missing something here.

America doubles down on computational fortresses while others optimize for effectiveness per watt, per chip, per dollar spent.

The irony cuts deeper when you consider we might be building the most expensive solution to problems that could have cheaper answers.

But then again, maybe scale is the answer. Maybe bigger really is better. Maybe I’m overthinking this.

When circular becomes routine

What strikes me most isn’t the scale—it’s how quickly we’ve adapted to it.

$100 billion commitments feel ordinary now. Stock movements measured in hundreds of billions based on what could be self-reinforcing financial flows.

Companies investing in their own customer base, then pointing to that spending as market validation.

Five years ago, this would have triggered immediate skepticism. Today, it’s Tuesday’s business news.

We’ve normalized financial arrangements that create their own gravitational pull. NVIDIA invests in companies that lease NVIDIA infrastructure. NVIDIA’s market value rises based on that “organic” demand. The cycle accelerates.

The shell game isn’t necessarily happening to us anymore.

The shell game might just be how things work now. Or call me naive if I’m too late to recognize this properly.

Lifting the lid on what might be engineered consensus

Here’s what this could mean for anyone trying to make sense of AI development.

When you see headlines about ChatGPT’s dominance or NVIDIA’s unprecedented demand, some of that success probably reflects genuine innovation. Some might reflect circular financing creating its own validation.

Separating signal from potentially engineered noise becomes harder when the same entities fund the demand they’re measuring.

The restaurant empire taught me something valuable: systems can work perfectly while being something entirely different from what they appear.

The food was still good. The service was real. Customers were genuinely satisfied.

But the competition might have been theater.

The question I keep coming back to: does it matter if the results are real?

Reality check time

NVIDIA’s $100 billion OpenAI commitment represents the largest AI infrastructure investment in history.

It also represents a financing model where companies invest in their own demand, creating market signals they then cite as validation.

The infrastructure going online in Abilene will be genuinely impressive. Whether it builds better AI or just more expensive AI—that’s the question I can’t shake.

Sometimes the most sophisticated solution is just the most profitable one for the people building it.

The shell game isn’t necessarily a form of deception.

It might just be architecture.

But I could be completely wrong about all of this, or naive.

Stay curious,

Mark
The AI Learning Guy
👋⚡😎

The AI Learning Guy newsletter 🤖 🧠💡

AI learning hacks and mega prompts delivered to your inbox.

Sources and books

Note: No single website has all the answers. This list serves as a starting point for those who want to explore or satisfy their curiosity about AI. Links: Links with * are affiliate links. See disclosure below.