Financial institutions are growing hesitant to carry AI debt liabilities, and that list includes the banks financing the Google-sponsored Texas data center under lease to Anthropic.

According to insiders, the project’s lenders, including Morgan Stanley, intend to offload their debt commitments for the 2,000-acre Texas data center campus by selling bonds as soon as the loans are drawn. Ideally, shedding the debt reduces AI risk and frees up vital capital for these lenders. 

The shift is part of the wider change in how Wall Street is financing AI infrastructure. Lenders are eager to participate in the growth of the sector but fewer and fewer are willing to hold large AI-related loans on their own balance sheets.

Data center developments require billions of dollars in upfront investment, long construction timelines, and depend on sustained demand from AI companies whose long-term revenue prospects remain uncertain. Selling the debt through bond markets allows banks to recycle capital while limiting exposure to a single high-risk sector.

Inside Google’s AI financing playbook

The bank-focused infrastructure market is already choking under the massive weight of AI funding needs. Primarily, construction, energy, and other infrastructure costs run well into the billions of dollars. For starters, Google had to anchor its Texas Campus project with an extensive network of deals and a financial program of over $150 billion, spanning from hardware manufacturing to data center construction. 

The building itself is financed by the main bank’s debt package, though the Google custom TPU chips will be financed on a separate track.

The financial structure relies on a syndicate including Broadcom, Apollo, Blackstone, and Morgan Stanley to backstop the core debt package. If Anthropic fails and the chips fall off in value, Broadcom steps in to cover the gap. Apollo and Blackstone also extend private credit via an SPV that leases the hardware to Anthropic, with Morgan Stanley as the financial adviser and lender of the deal.

Speaking on their arrangement, A Google executive noted, “This is each of us putting our balance sheet to work. We’re doing it on the data center side, [Broadcom’s] doing it on the chip side.”

With all the arrangements, the firm still expects about a 20% equity stake in the Hubbard, Texas, campus, according to sources.

Nonetheless, it intends to use an on-site natural gas plant to eliminate grid delays and reduce energy costs. The sheer volume of upcoming data center builds in Texas, however, is already sparking fears over grid capacity, water shortages, and spiking utility rates.

Furthermore, fusing data center and power assets complicates funding structures, as lenders must evaluate two distinct risk profiles simultaneously. In the case of Project Walleye, a parallel Meta-data hub running on its own dedicated energy supply, lenders demanded—and received—inflated yields to underwrite the dual-asset risk.

Banks may issue bond sales to offload debt

For the Texas campus, sources familiar with the transaction say, however, the $15 billion debt will be carved into multiple bond sales to mirror a delayed-draw feature, allowing developer Nexus Data Centers to pull down funds sequentially as construction hits designated milestones. 

Some of the financing might even be repurposed using leveraged loans. The bond could also have a speculative-grade rating as Google’s backstop is still in limbo until the data center is fully built, analysts say. Investors would also face the risk of construction delays and budget overruns, the structure says.

But investors will remain interested and that is because the project is backed by companies like Google and Anthropic. Infrastructure-linked assets have attracted strong demand as they are often more attractive than investment-grade debt. However, pricing will still take into account construction risk and uncertainty in the timeline of the project before Google’s guarantees actually come into effect.

More recently, the bond market has become a preferred source for AI projects to raise long-term funds in a faster and less expensive way than bank loans.

Meanwhile, in the past few months, banks have also aggressively tried to unload $50 billion in Oracle infrastructure loans, turning to risk-transfer markets to shield themselves from exposure.

Analysts say the outcome of the Texas financing could influence how future AI infrastructure projects are funded. If investors absorb the bonds without demanding significantly higher yields, other technology companies may adopt similar financing structures for multi-billion-dollar data center developments.

Conversely, weak demand or wider credit spreads could make borrowing more expensive and reshape how banks participate in the AI infrastructure boom.

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It’s free.