Financing the AI Infrastructure Era

Originally published in July 2026, building on an earlier note on AI data-centre financing.

The AI race is not only about models.

It is increasingly about securing the capital, chips, data-centre capacity and power required to train and operate them at scale.

That makes AI infrastructure not only a technology theme, but a financing theme.

A multi-trillion-dollar buildout

Industry estimates cited by Apollo put required AI infrastructure investment through 2028 at close to $3 trillion.

Hyperscaler cash generation can fund a large part of that bill, but not all of it. Estimates suggest roughly $1.5 trillion may need to come from external capital, creating a financing requirement too large for any single market to absorb alone.

Public bonds, banks, project finance, joint ventures, securitised structures and private credit can all participate.

Apollo has estimated that private credit — particularly asset-backed finance — could provide more than $800 billion of the required external capital.

Why private credit fits

AI infrastructure contains characteristics that can be attractive to private lenders.

Data centres are physical assets. Compute equipment has identifiable value. Long-term contracts with large technology companies can create predictable cash flows. Power infrastructure and real estate can support additional forms of collateral.

But those characteristics should not be confused with low risk.

The underwriting question is not simply whether AI demand will grow. It is whether the specific asset, contract and capital structure remain money-good through technological and economic change.

The risk behind the boom

The scale of announced investment introduces several risks.

First is utilisation. Capacity built on aggressive demand assumptions may not operate at the expected level.

Second is technology obsolescence. GPUs and related equipment can lose economic value much faster than traditional infrastructure assets.

Third is counterparty concentration. A project may ultimately depend on a small number of hyperscalers, AI labs or tenants.

Fourth is power. Data centres are unusually energy intensive, making grid access, generation capacity and permitting central to both project economics and completion risk.

Finally there is residual value. A lender relying on collateral must ask what that collateral is worth if the original business plan fails.

From corporate credit to asset-backed finance

The financing need is also broadening the boundary between traditional corporate lending and asset-backed finance.

Instead of lending solely against a company’s enterprise value, capital can be structured around individual data centres, equipment pools, contracted cash flows or special-purpose vehicles.

That can isolate assets and create more tailored risk allocation.

But structural complexity does not eliminate economic risk. It redistributes it.

What happens if expectations are wrong?

This is where the credit perspective differs from the equity narrative.

An equity investor can justify a high valuation through enormous upside if AI transforms productivity and creates new markets.

A lender does not receive that same upside.

The lender needs to know that principal and interest can be repaid even if adoption is slower, pricing falls, hardware becomes obsolete or refinancing conditions tighten.

That means stressing utilisation, lease renewal, counterparty quality, equipment values, power costs and take-out assumptions.

An investment perspective

The AI infrastructure boom may become one of the largest capital-formation events of this decade.

That does not automatically make it one of the best credit opportunities.

The scale of external financing required means private markets are likely to play an important role. But the most interesting question is not how much capital AI needs.

It is which risks lenders are being paid to absorb — and whether the structure protects them if technological optimism outruns cash generation.

The technology can be transformative while individual financings still disappoint. Credit analysis has to hold both ideas at the same time.