US chipmaker NVIDIA has announced a major partnership with several of the world's leading financial institutions and asset managers. Together with partners including BlackRock, Blackstone, Apollo, Brookfield, Goldman Sachs and KKR, the company aims to establish independent financing platforms. The stated goal of this initiative is to mobilize over 500 billion US dollars in third-party capital for the expansion of global AI compute infrastructure. This massive move underscores the extraordinary capital requirement that the global rollout of artificial intelligence now demands.
The announcement comes at a time when spending on artificial intelligence infrastructure is acquiring macro-level economic significance. According to forecasts by J.P. Morgan, the capital expenditures of top hyperscalers are expected to reach 697 billion US dollars in 2026. An analysis by LSEG Data and Analytics even places the projected capital spending of the five largest US hyperscalers at 720 billion US dollars for 2026. These unprecedented expenditure sums are increasingly influencing global capital markets and interest rate structures worldwide.
At the same time, operational hurdles and execution risks for physical data center projects are growing rapidly. Reports indicate that major lending consortia, including JPMorgan, Bank of America and Morgan Stanley, are applying significantly stricter due diligence standards for project loans. Drivers for this enhanced scrutiny include rising local community opposition, environmental regulations and severe power grid bottlenecks. Lenders are becoming increasingly cautious as potential project delays threaten completion schedules and financial returns.
Beyond project risks, rating agencies and regulatory authorities are warning about systemic concentration in the financial sector. Rating agency Moody's recently highlighted the risks for banks relying on a small group of tech giants for basic AI models and cloud infrastructure. Beyond vendor lock-in and potential outages, analysts point to risks such as automated deposit flights and heightened cybersecurity vulnerabilities. These concerns underscore the urgent need for well-structured and diversified infrastructure financing.
Meanwhile, the underlying cost dynamics of operating artificial intelligence are undergoing a fundamental shift. Gartner forecasts reveal that enterprise spending on executing AI models will exceed pure model training expenditures in 2026. Estimated inference costs of 23.3 billion US dollars will surpass training spending of 19.0 billion US dollars as agentic software deployment accelerates. The shift toward continuous real-time execution in enterprise workflows illustrates why massive infrastructure funding is essential to sustain long-term capacity.