BlackRock’s Larry Fink says AI data centres need a new way to raise capital

The AI data-centre build-out is running into a financing question as well as a construction one: who can fund infrastructure on this scale, and how? BlackRock chief executive Larry Fink says capital is already tight for bespoke data-centre deals, while arguing that better data and standardised risk measures could help bring more investors into the market.

Watch Desk analysis

What happened

In a Diginomica article dated 16 October, Stuart Lauchlan reports Fink’s remarks on the financing challenge. Fink compares compute infrastructure with the early mortgage-backed securities market, arguing that trusted performance data and common risk measures helped turn a narrow pool of mortgage loans into a much larger investable market.

Fink says a one-gigawatt US data centre can cost about $50 billion, with GPUs and chips accounting for roughly 40 per cent and power, grid, core and shell making up the rest. He also puts US needs for grid, power and data-centre construction at $7 trillion to $9 trillion over several years. Those are Fink’s estimates, not independently established project totals.

Why it matters

The figures make clear why financing is becoming part of the AI infrastructure story. If compute capacity is to expand beyond the largest technology companies, lenders and investors need ways to assess projects and their risks, not just enthusiasm for AI demand.

Fink’s proposed answer also puts BlackRock in the picture as a potential organiser of capital, not merely a spectator. That is an opportunity for the asset manager, and a reminder that the rules and data used to price this infrastructure will matter to who can build it.

Our read

The useful point is not that AI investment is doomed or guaranteed to pay off. It is that the build-out depends on financing structures capable of carrying enormous costs, and those structures are still being worked out. Fink has a commercial stake in that conversation, so treat his forecasts as an interested view, not a neutral balance sheet of the future.

What to watch

• Whether investors and lenders adopt common performance and risk measures for compute infrastructure.
• How much of the proposed build-out is financed through new structures rather than conventional borrowing.
• Whether access to financing broadens beyond the biggest technology companies.

Discussion spark: Should financing for AI infrastructure be built around standardised performance data, or does that risk making a fast-moving market look more predictable than it is?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.