When Jensen Huang writes a $2 billion check, you pay attention. But when that check goes to a company that primarily rents out Nvidia's own chips, you start asking questions about who's really propping up demand.

Nvidia announced Monday it's investing $2 billion in CoreWeave, an AI infrastructure provider that generates revenue by building data centers packed with—you guessed it—Nvidia's graphics processing units. CoreWeave's stock jumped 8% in premarket trading, though that's from a share price ($87.20) that's actually below Friday's close of $92.98.

Here's the pitch: CoreWeave is racing to build "5 gigawatts of AI factories by 2030." To put that in perspective, five gigawatts is roughly the annual power consumption of 4 million U.S. households. That's not a data center—that's a small country's worth of electricity dedicated to training AI models.

"CoreWeave's deep AI factory expertise, platform software, and unmatched execution velocity are recognized across the industry," Huang said in a statement. "Together, we're racing to meet extraordinary demand for NVIDIA AI factories—the foundation of the AI industrial revolution."

But let's talk about the elephant in the data center: Nvidia already has a massive stake in CoreWeave's success. Back in September, CoreWeave disclosed an order worth at least $6.3 billion from Nvidia, with Nvidia obligated to buy "residual unsold capacity" through April 2032. Now Nvidia is also an equity investor.

This creates what finance people politely call "circular dynamics" and skeptics call "artificial demand." Nvidia makes the chips, invests in the companies buying the chips, and guarantees to buy unused capacity from those companies. It's vertical integration meets financial engineering, and it raises a legitimate question: is CoreWeave building infrastructure because customers need it, or because Nvidia needs someone to buy its chips?

To be fair, CoreWeave has real customers. The company announced contracts with Meta worth $14.2 billion and OpenAI for $22.4 billion. Those are genuine AI infrastructure needs from companies with genuine AI products.

But here's where it gets tricky for investors: CoreWeave went public in March and has been on a deal-making blitz ever since, raising billions in debt and equity. According to CNBC, the company primarily generates revenue by renting out Nvidia GPUs—which means its business model is entirely dependent on one supplier who is also now a major shareholder.

For those keeping score at home, Nvidia's valuation depends on insatiable demand for AI chips. CoreWeave's valuation depends on being able to rent out those chips. And now Nvidia has guaranteed to backstop CoreWeave's capacity through 2032 while also taking an equity position.

Is this a problem? Maybe not. AI infrastructure needs are real, and someone has to build the "factories" Huang keeps talking about. But when your biggest customer is also your supplier, investor, and backstop buyer, that's not diversification—that's concentration risk dressed up as growth.

For investors in either company, the question isn't whether AI infrastructure is important. It clearly is. The question is whether these intertwined relationships represent genuine market demand or a carefully constructed ecosystem designed to keep demand looking strong.

Time will tell. But when the music stops and someone needs to pay for all those gigawatts of capacity, the bills will be very, very real.