The CEO of the world’s largest asset manager said something that reframes how investors think about the AI trade. Larry Fink, managing $11.5 trillion at BlackRock, said at the Milken Institute Global Conference: “We just don’t have enough compute.”
Fink stated: “The United States is short power. We’re short compute. We’re short chips. And there’s going to be shortages in all three and memory, four things.”
He added: “I actually believe a new asset class will be buying futures of compute.”
Compute as a Tradable Commodity
Fink is predicting compute becomes a tradable commodity like oil or natural gas, with derivatives markets emerging to price it. He defined compute not merely as a cloud service but as a scarce resource that financial markets can price — comparing it to commodities like energy or agricultural products. Companies could manage AI infrastructure costs through structured futures contracts, similar to how they hedge fuel prices.
“The new asset class is buying compute futures.” This is viewed as a plan to bundle physical infrastructure — data centres, semiconductors, memory, and power — into underlying assets for financial products as AI investment grows. For institutional investors, it could provide a mechanism to lock in costs for the chip and power capacity required to run models.
Supporting Data Points
- Data centers will consume 70% of all memory chips globally in 2026
- HBM production from Samsung, SK Hynix, and Micron is sold out through 2026-2027
- A single AI server uses 10-20x more memory than conventional servers
- DRAM supply grows at 16% annually while AI infrastructure demand grows at 80%+
Not a Bubble — Supply Shortages
Fink also stated: “There is not an AI bubble. There is the opposite. We have supply shortages. Demand is growing much faster than anyone has ever anticipated.”
He emphasized that current supply conditions are falling behind AI demand. The U.S. lacks sufficient chips, memory, and power facilities for expected AI workloads. Even though cloud companies and large semiconductor firms are raising substantial funds, that alone may fall short of meeting global demand for expanding data centres. He cautioned that the sector could face a scenario where capital itself becomes scarce.
Infrastructure Investment Scale
- Data center costs: $50-75 billion for a one-gigawatt data center
- Brookfield CEO Bruce Flatt estimated $10 trillion needed for power, AI factories, data centers, and fiber build-out to “rewire the world”
- Fink said private capital (pension funds, 401(k)s, sovereign wealth funds, insurance companies) would play a central role, much of it financed by debt and private credit
BlackRock Partnerships
Fink hinted at an upcoming partnership with an unnamed hyperscaler to build data centers, to be announced later that week. BlackRock (40 billion Aligned Data Centers deal.
Standardization Challenges
Before a compute futures market can be created, standardization issues must be resolved. Exchanges listing such products would need to decide on a standard unit, accounting for differences in hardware across generations and constantly shifting AI workloads. The industry has yet to reach consensus on how to measure computing performance and convert it into contract units.
The key open questions: whether the industry can establish a standard unit for compute and whether a derivatives market will open based on it.
Original: MilkRoadAI @ X