Tiramisù or AI Data Centers? A sweet bet that can turn bitter!

Remember NVIDIA’s supposedly $500 billion AI infrastructure announcement?

ℹ️️On August 2026, NVIDIA partnered with Apollo Global Management, Inc., BlackRock, Blackstonee, Brookfield, Goldman Sachs and KKR to establish AI Compute Infrastructure Financing Platforms to Mobilize Over $500 Billion of Third-Party Capital !

⚠️According to recent Bloomberg reporting, cited by Ed Zitron, the deal’s financial origins were unclear, while financing negotiations for data centers were progressing slowly. Yet the headline travelled around the world, inflating the bubble!

⚠️Meanwhile, Goldman Sachs is warning about credit risks in data-center financing: construction costs, refinancing, electricity availability, permits and insurance.

⚠️Real-world constraints are finally catching up with the AI hype.

🚩And then came CoreWeave. According to "A Tale of Two Bubbles" that great post by Ed Elson, its unsecured bond yields reportedly surged to 15%, while credit-default swaps implied a 62% probability of default within five years. The article describes a company carrying approximately $50 billion in debt at around 9% interest while generating a return of just 1%. Borrow expensive money to earn less than the cost of borrowing? What could possibly go wrong?

🚩But the biggest bubble might not be where everyone expects. The real danger may be the private valuations of OpenAI and Anthropic. Elson argues that these AI laboratories are being valued at extraordinary levels despite reportedly losing billions annually and facing unresolved questions about the sustainability of their business models. Private-market valuations are not the same as valuations determined by public-market investors demanding profits.

ℹ️️Now connect the dots.

NVIDIA reportedly derives approximately 75% of its revenue from just five customers. Among the key players in this ecosystem are OpenAI and Anthropicc, whose underlying financial sustainability remains difficult for outsiders to assess fully.

🚩Oracle, meanwhile, is making enormous infrastructure commitments tied to AI demand, including business linked to OpenAI.

If AI laboratories cannot justify their valuations, financing becomes more expensive, or customers cut spending, the consequences could spread across chipmakers, data-center operators, lenders and infrastructure investors.

📍One interconnected ecosystem.

📍Multiple points of failure.

ℹ️️Of course, AI is real, and its long-term potential is enormous. But technological revolutions do not immunize investors against overvaluation, excessive leverage or bad economics.

🚩The next couple of months could deliver some very uncomfortable reality checks.

🎁 I’ve already bet my friends a tiramisù that the AI bubble will burst soon.

🍰 If I’m wrong, dessert is on me.

🥒 If I’m right, the whole industry may end up with a much bigger bill, and a very bitter aftertaste !