Nvidia chief Huang sees AI infrastructure emerging as a new asset class

In recent quarters, several major asset managers and banks have announced funds and vehicles aimed specifically at financing AI data centers. Named in this context are, among others, BlackRock, Blackstone and Goldman Sachs, which together with tech companies such as Amazon and Microsoft are investing billions in sites and power grids.

Observers see parallels to the financing of the US real estate market before the 2008 financial crisis. Back then, mortgage loans were bundled into securities and sold worldwide. Critics now ask whether the rapidly growing AI data centers could form a similar bubble – a comparison Huang himself does not draw.

Risks and parallels to the real estate bubble

Supporters argue that demand for AI services has been growing steadily since the breakthrough of generative models and that companies such as Nvidia, Amazon and Microsoft have already signed long-term offtake agreements with cloud customers. These contracts would provide stable revenues and make the investments less risky than pure speculation.

At the same time, analysts point to risks: the power supply for large data centers is tight in many regions, permitting procedures sometimes take years, and individual providers are heavily dependent on a few major customers. Should demand for AI services grow more slowly than expected, newly built capacity could sit idle.

Huang's remarks stirred movement on Wall Street on Tuesday. Nvidia shares rose at times in pre-market trading, while stocks of some data center operators and utilities also gained. Traders cited growing interest from institutional investors in the topic.

Reactions on Wall Street

Market observers stress that "AI infrastructure" is currently not a uniformly defined term. It usually refers to large-scale data centers with thousands of specialized chips, fiber-optic connections and their own power supply – facilities often referred to as "AI factories."

Whether this trend will develop into a new open capital market in the style of the bond market depends, according to experts, primarily on the willingness of large asset managers to create standardized products. Initial vehicles from BlackRock and Blackstone are said to already give institutional investors access to individual sites or entire portfolios.

Industry insiders nevertheless see a difference from the subprime crisis of that era: today's underlying investments are more tightly tied to concrete usage agreements with technology companies. However, broad, transparent price discovery – which would let investors truly gauge how risky the commitments are – is still lacking.

How close is the similarity to subprime?

In his speech, Huang also called for closer cooperation between chip manufacturers, cloud providers and the financial industry. Only if capital flows quickly and at scale into sites, networks and power, he said, can the next wave of AI applications become available in time.

The term "open capital market" is defined in finance as a market in which companies raise capital through standardized, regulated instruments such as bonds or stocks. Whether such a market will actually develop for AI infrastructure remains an open question.