Nvidia’s new financial strategy does not compute

Nvidia CEO Jensen Huang just announced a $500 billion partnership with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to establish compute as an investable asset class, according to The Verge. Huang told CNBC this marks the first time technology chips have become investable assets, calling them "revenue-generating, productive, long-lived, fungible, and flexible."

Here's the problem with that pitch. Traditional asset classes like real estate, commodities, or bonds have stable, predictable value curves. A GPU's productive lifespan is measured in months, not decades. The H100 chips everyone scrambled to buy 18 months ago are already being displaced by newer architectures, and AI model efficiency improvements are cutting compute requirements faster than most anticipated.

The financial engineering here is familiar. Wall Street loves turning things into tradable securities, from mortgage-backed obligations to weather derivatives. But compute has a depreciation curve more like smartphones than buildings. By the time these investment vehicles mature, the underlying hardware may be training-obsolete, useful only for inference on legacy models.

Nvidia benefits enormously from this arrangement regardless of whether it works long term for investors. It creates massive guaranteed demand for their chips, offloads risk to financial institutions, and positions compute capacity as infrastructure rather than technology expenditure. It's brilliant for Nvidia's balance sheet.

The broader implication is that AI infrastructure is now being built on Wall Street's timeline and risk appetite, not on technical fundamentals. That creates interesting distortions. Expect more data centers, more capacity, and possibly more computing power than the market can efficiently use, at least in the near term.

What this means for you: If you're building AI products or services, cheap inference capacity may become more available as these financial vehicles need to generate returns. Start thinking about how you'd scale if compute costs dropped 50% or more. Try this prompt with your AI assistant: "Help me redesign our product architecture assuming inference costs fall to near zero over the next 18 months. What features become viable? What would we build differently?" The financialization of compute may create opportunities for builders even if it burns investors.