Major Wall Street financial institutions are pushing plans to transfer $15 billion of loans for a Texas artificial intelligence (AI) data center project backed by Google to the bond market.

On Aug. 5 local time, blockchain media outlet Cryptopolitan reported that a lending group including Morgan Stanley is reviewing a plan to move to sell bonds immediately if the loans are actually funded.

The deal centers on a 2,000-acre data center campus in Hubbard, Texas, structured for Anthropic to lease. Banks are joining the growth of AI infrastructure, but are becoming increasingly cautious about holding related loans on their own balance sheets for long periods. Data center development requires initial investment costs of several billion dollars and has long construction periods. The added burden is that long-term demand depends on the performance of AI companies. Sales through the bond market allow banks to recycle capital while limiting exposure to specific high-risk sectors.

The banking sector's infrastructure finance market is already being weighed down by AI funding demand. Infrastructure costs, including construction and energy, have swelled into the billions of dollars. Google is reported to have put together a financial programme and contract network exceeding $150 billion for this project, from hardware production to data center construction.

The buildings are financed with a core bank debt package, while Google's customised TPU chips are funded under a separate structure. Broadcom, Apollo, Blackstone and Morgan Stanley are involved in the transaction structure. Broadcom takes on a role of covering losses if Anthropic fails and chip values fall. Apollo and Blackstone provide private credit that leases equipment to Anthropic through special purpose vehicles. Morgan Stanley is handling financial advice and the loans.

Google described the structure as "each using its own balance sheet" and said, "We handle the data center side and Broadcom handles the chip side." Google is expected to hold about a 20 percent stake in the Hubbard campus even under this structure.

Google plans to run a natural gas power plant on the site to avoid delays in connecting to the power grid and to lower energy costs. Still, a string of data center construction in Texas is increasing concerns about power grid capacity, water shortages and surging electricity bills.

If data centers and power assets are combined, the lending group must bear the burden of assessing two types of risk at the same time. In Meta's similar project, Project Walleye, the lenders are reported to have demanded higher returns citing this dual-asset risk.

For the Texas campus deal, the $15 billion debt is set to be split into multiple bond issues to reflect construction-stage drawdowns, with developer Nexus Data Centers pulling in needed funds sequentially each time it meets construction milestones. Some funds may also be restructured as leveraged loans, it is said.

An analysis says Google's support commitments may only take full effect after the data center is completed, raising the possibility that the issued bonds could receive speculative-grade ratings. Investors would also have to bear risks of construction delays and cost overruns.

The fact that Google and Anthropic are behind the project is cited as a factor supporting investor demand. Infrastructure-linked assets can offer higher returns than investment-grade debt, and inflows are also being seen. But until Google's guarantee becomes effective, construction risk and schedule uncertainty are likely to be reflected in pricing.

The bond market has recently emerged as a channel for raising long-term funding for AI projects faster and more cheaply than bank loans. Banks are also reported to have actively reduced $50 billion of Oracle-related infrastructure loans in recent months by using the risk-transfer market.

Analysts see the outcome of the Texas deal as potentially affecting how future AI infrastructure projects are financed. If investors absorb these bonds without a large interest-rate add-on, other technology companies may adopt similar structures for multibillion-dollar data center development. If demand is weak or credit spreads widen further, AI infrastructure borrowing costs could rise and banks' participation could also change.