Fink: Futures on AI computing could become asset class

Speaking onstage at the Milken Institute conference in Beverly Hills, BlackRock CEO Larry Fink told attendees that rising demand for AI compute could support a new asset class based on futures for raw computing power.

Fink compared compute to energy and agricultural commodities that markets already hedge with forward agreements, and he suggested exchanges could eventually list standardized futures tied to capacity.

Those contracts would let buyers lock prices for delivery of power, processing hours or specific chip and memory configurations at a future date. Institutions financing AI build-outs could use such contracts to hedge capacity costs in a similar way airlines hedge fuel.

Fink said the United States does not yet have enough chips, memory or power capacity for projected AI workloads and that the sector could attract long-term investors seeking direct exposure to physical infrastructure. He stated, “A new asset class will be buying futures of compute.”

Rejecting the idea that AI investment is a bubble, he said demand still outpaces supply across the hardware and infrastructure stack and warned of a coming shortage of capital. He added, “We’re short power, we’re short compute, we’re short chips.”

BlackRock is preparing to announce a partnership with an unnamed hyperscaler that would move the firm into taking direct stakes in data centers and related physical assets. The deal would expand the firm’s role beyond financing; Fink declined to name the partner ahead of a formal announcement.

Whether exchanges adopt compute futures will hinge on agreeing a standard unit of compute across hardware generations and shifting AI workloads, plus mechanisms to verify delivery and capacity. That could include measures such as megawatts, processor hours, or defined chip and memory configurations.

Fink framed the potential contracts as a way to assign clear prices to the megawatts, chips and memory behind each model query and described how markets might use futures to manage price risk for compute.