Three Hidden Trends in AI Infrastructure Financing Are Twisting Together—Will They Strangle the Industry or Not?

In 2013, Deutsche Börse and Berlin-based cloud technology company Zimory formed a joint venture to create a "cloud exchange," aiming to turn computing power into a tradable commodity like oil. Three years later, the project was shut down—CPUs were too difficult to standardize, there weren't enough customers, and no one came to trade. Thirteen years later, the same idea has resurfaced, but with a completely different background: in the first quarter of 2026, global enterprises spent $129 billion on cloud services, about 15 times the amount in the same period a decade ago. This year, the combined capital expenditure of the top five U.S. hyperscale companies is expected to approach $700 billion. The term "computing power" has moved from being a technical term to a regular feature in financial headlines.

Surrounding these astronomical figures, an entire financial infrastructure is being built: GPUs are being used as collateral for loans, computing power rental prices are formed into indexes, CME and ICE are racing to launch computing power futures, and ETF issuers are already lining up with applications—even some are working on 2x leveraged funds for computing power. This set of activities is collectively known as: the financialization of computing power.

Is this a sign of a market maturing, or is it adding another layer of leverage to an industry that's already addicted to burning cash?

To answer this question, we can't just focus on those futures contracts that haven't really been traded yet. What truly deserves concern is a deeper structure—it's twisted together by three strands, each of which alone may not seem very risky, but when intertwined, forms the most fragile knot in this round of AI infrastructure cycle.