AI’s power crunch is driving a multibillion-dollar bank rush

The AI infrastructure bottleneck has shifted from Nvidia GPUs to electricity, and data-center builders are turning to banks for billions of dollars in financing commitments before they sign power and construction contracts.

As TNW reports, citing Bloomberg, the four largest cloud companies are expected to spend about $725 billion on AI infrastructure this year. Meta’s planned Texas data center alone carries a $13 billion price tag—an investment that would once have funded an entire company.

The underlying problem is density. Server racks used for conventional computing drew roughly 3kW; AI racks can consume up to 150kW. Global data-center capacity is expected to nearly double to about 200GW by 2030, creating demand for electricity, substations, long-term power contracts, and the financing to secure them.

That pressure is already driving efforts to develop alternative energy sources for AI data centers, including the fusion and microreactor projects pitched to Microsoft — though the source does not establish that these projects are connected to the financing activity described here.

Recommended reading

Apple asks court to block OpenAI over alleged trade secrets

Banks finance the power race

Builders increasingly want firm lending commitments before committing to electricity deals or construction. The sums are too large for many companies to carry on their own balance sheets, pushing lenders toward a mix of conventional loans, bonds, and private-credit arrangements.

The deals cited by TNW include:

• A $5.9 billion loan for one data-center operator.
• Oracle’s $16.3 billion data-center raise, which relied on private credit after banks became more cautious.
• Around $1.65 trillion in estimated off-balance-sheet obligations, structured through special-purpose vehicles that keep debt away from major technology companies' own accounts.

The increasingly complex financing is attracting regulatory scrutiny. The Bank for International Settlements has identified “circular financing” between cloud providers, suppliers, and construction lessors as a significant potential financial-stability risk. The concern is that the same growth assumptions can support multiple layers of borrowing, making the system harder to unwind if AI demand slows.

Data centers reshape energy and construction

The rush for power is changing energy markets as well as credit markets. Developers are reviving gas plants, signing nuclear-power agreements, and moving ahead in utility queues. Data centers are becoming some of the largest new electricity buyers in a generation, while strained grids are already contributing to higher household costs in some regions.

Construction capacity is another constraint. Fewer than 10 companies worldwide can deliver hyperscale projects, and Turner reportedly has a record backlog of about $44 billion, much of it tied to data centers.

Banks are not treating the boom as risk-free. Some have pulled back as financing structures become larger and more circular, which helps explain why developers are seeking firm pledges instead of assuming capital will remain available.

The entire buildout depends on AI usage continuing to rise quickly enough to fill the new capacity. Goldman Sachs expects cumulative spending on the expansion to reach the trillions by 2030. That makes the financing surge both an enabler and a vulnerability: the more capital committed to power and infrastructure, the more urgently the industry must deliver the AI revenue needed to justify it.