Larry Fink Says the U.S. May Need More Than 70 Gigawatts of Power to Run AI

BlackRock CEO Larry Fink said the United States will need “over 70 gigawatts of power” to support the artificial intelligence boom, framing electricity supply as a central constraint for the next phase of AI growth. Speaking on a CNBC panel that also referenced NVIDIA, Fink positioned the power requirement as an estimate of what the U.S. alone may need, rather than a global figure.

Gigawatts measure the scale of electricity capacity, with 1 gigawatt equal to one billion watts. In the context of AI, the figure underscores a shift in how investors and corporate leaders are thinking about computing: not only about chips and software, but about the grid capacity required to feed data centers and related infrastructure.

The remarks land as power demand has become a recurring topic in AI-sector discussions, because training and running AI workloads can draw substantial and increasingly predictable energy loads. While the underlying driver is computational demand, the bottleneck often moves to electricity availability, permitting, and timelines for new generation and transmission.

As a major asset manager, BlackRock’s business depends on long-term capital allocation across equities, bonds, and infrastructure-adjacent themes. Fink’s comment puts the investment conversation directly on energy buildout needs, suggesting that AI’s growth curve may be shaped in part by whether the U.S. can add enough power capacity to match demand.

The CNBC discussion tied Fink’s estimate to the broader AI supply chain, including NVIDIA, which is widely associated with advanced AI processing hardware. The announcement behind this story does not provide additional details about NVIDIA’s role in the power calculation, but the pairing reflects how investors commonly connect AI compute to downstream operational requirements such as data center energy and grid expansion.

Even if “over 70 gigawatts” is an estimate rather than a formal forecast, it reflects a clear managerial point: large technology spending cycles can collide with physical infrastructure limits. For companies building, expanding, or financing data centers and power systems, the practical question is less about whether AI will grow and more about how quickly energy capacity can be delivered at the needed locations.

The post also does not disclose whether BlackRock has adopted a specific policy or investment product tied to that number, nor does it provide a breakdown of how the total would be allocated across data centers, supporting grid upgrades, or other energy-intensive uses. It similarly does not explain whether the estimate assumes particular technology efficiency improvements, regulatory timelines, or specific AI workload profiles.

Going forward, investors will likely watch for additional specificity around power demand estimates and for corporate commentary that links those figures to capex plans, utility buildouts, and data center development schedules. Another near-term indicator may be whether more AI and infrastructure leaders discuss power capacity targets with similar precision, and how those targets translate into procurement and construction timing.