AI data center financing has passed $1.8 trillion. Companies building the facilities raised that sum across debt, equity and securitized instruments between 2020 and the first half of 2026,. Nearly half of it arrived in just three quarters.
Behind the headline is a less comfortable number. At least $1 trillion of the debt matures within the next five years, and lenders are already pricing in more credit risk.
The pace of fundraising
Developers raised $794 billion between the third quarter of 2025 and the first half of 2026. That is close to half of the total raised since 2020.
Private markets are taking a larger role. They supplied 56% of colocation financing in the first half of 2026. Colocation providers, as a group, have raised more than the four hyperscalers combined.
Why refinancing is a problem
Several pressures are stacking up. Interest rates are higher, worries about developers’ debt levels are growing, and many firms lean heavily on support from hyperscalers and chipmakers. Together they make rolling over old loans harder.
Colocation providers and neoclouds face the steepest climb. Neocloud debt maturing between 2029 and 2030 is 18 times what it was a year earlier. Smaller developers such as TeraWulf and Crusoe carry added uncertainty because of large construction loans.
Execution problems have already hurt two companies. Firmus and Nscale both saw their high-valuation IPO plans affected.
How the deals are built
Deal structures are getting more complex. Smaller developers often can’t raise money at the size and tenor they need on their own, so they borrow against the credit of investment-grade backers.
- GPU-backed loans
- Direct guarantees
- Residual-value guarantees
- Prepayments
Nvidia has extended at least $198 billion in direct credit support and has committed more alongside large asset managers. That ties the chipmaker’s balance sheet to its customers’ ability to repay.
Which banks are leading
Morgan Stanley underwrote the most data center financing, helped by its close relationship with Meta. Citigroup and JPMorgan Chase follow. TD Bank and Scotiabank sped up dealmaking the most in 2026 as they competed to fund established players.
How the analysis was done
The research compiled more than 2,000 transactions from the 125 largest data center developers. It drew on the Bloomberg Terminal, third-party data and news reports, and added industry interviews and document reviews for sector and deal detail.
Why the mix of lenders matters
Hyperscalers are the most obvious builders, but the colocation providers that lease space to them have raised more money. That puts a large share of the risk on companies with far thinner balance sheets than the tech giants they serve.
Neoclouds, which rent out GPU capacity, sit in a similar spot. Their debt schedules bunch up in 2029 and 2030, so any slip in demand for AI compute would land on them at the same time as their repayments.
What it means for the AI buildout
The spending isn’t slowing, but the way it is funded is changing. Early projects leaned on the cash piles of a few giant tech companies. Newer ones depend on a wider mix of lenders, and on guarantees that move risk onto Nvidia and the hyperscalers.
That makes the system more sensitive to demand. If AI revenue grows as developers expect, the debt gets repaid from leases. If it doesn’t, guarantors could be asked to cover shortfalls, and AI data center financing could tighten for everyone.
Smaller names are the likeliest to feel it first. Those with large construction debt and few investment-grade tenants will need to refinance in a market that now charges more for risk.
What to watch through 2030
The key date range is 2029 to 2030, when neocloud maturities jump. Watch whether private credit keeps its 56% share of colocation deals, whether more IPOs stall the way Firmus and Nscale did, and whether Nvidia’s $198 billion in credit support keeps growing. How those three play out will decide if AI data center financing at this scale is sustainable or merely early.
Source: BloombergNEF