Wall Street’s biggest banks are preparing to hand off $15 billion in construction debt tied to a sprawling Google-backed data center in Texas. Anthropic has leased the site, and that lease now sits at the center of one of the largest AI infrastructure financing deals of the year. Morgan Stanley leads a consortium of lenders that plans to tap the bond market once borrowers draw the loans. The move signals a broader shift in how banks manage exposure to artificial intelligence infrastructure projects. Lenders no longer want this kind of debt parked on their books for long. They want to sell it into capital markets deep enough to absorb enormous sums, and fast.

The project at the heart of this deal spans 2,000 acres in Hubbard, Texas, where construction crews are racing to build one of the largest data center campuses in the country. Nexus Data Centers developed the site, and Google has agreed to back it financially, giving the project the kind of institutional credibility that draws bond investors. Banks committed the initial $15 billion in construction financing, but they never intended to hold that debt to maturity. Instead, they structured the deal so they could refinance it in public markets as soon as Nexus Data Centers draws the funds. That timing detail matters enormously to how the deal will play out over the coming months. Bankers familiar with the plan say they will likely divide the debt into multiple bond offerings instead of marketing it as one enormous package.

Why Wall Street Is Rushing to the Bond Market

Banks have discovered that bond markets offer something bank loans cannot match at this scale: depth. AI infrastructure projects now require tens of billions of dollars apiece, and that kind of capital simply outstrips what banks can comfortably underwrite and hold. Selling debt into the bond market gives large infrastructure projects access to a broader pool of long-term investors, allowing banks to recycle capital after they commit construction financing. For the banks themselves, offloading this debt also reduces their overall exposure to AI infrastructure risk, freeing up balance sheet capacity for future lending. As a result, banks can keep participating in the AI buildout without absorbing the full weight of every project they help finance. AI infrastructure financing has become one of the largest new sources of demand in project finance, joining sectors such as energy, transportation, and telecommunications that have traditionally relied on large-scale capital markets.

The $15 billion loan carries what bankers call a delay-draw feature, a structure that allows Nexus Data Centers to withdraw funds gradually as construction hits specific milestones. This mechanism protects lenders from releasing all their capital upfront on a project that is still years from completion. It also explains why the refinancing timeline remains flexible instead of following a single closing date. Bond sales will likely follow the pace of construction draws, meaning the consortium could bring new offerings to market in stages over an extended period. The consortium may also refinance part of the debt package through the leveraged loan market instead of relying solely on bonds, giving it additional flexibility. Morgan Stanley and Nexus have both declined to comment on the specifics of the arrangement.

Oracle’s Shadow Looms Over the AI Debt Pipeline

This Anthropic-linked deal does not exist in isolation, and bankers on Wall Street have spent months already working through a much larger backlog. Major banks have been searching for buyers for more than $50 billion of construction debt tied to several data center projects that Oracle leased earlier this year. Some lenders managing Oracle exposure have turned to risk-transfer deals to offload portions of their holdings without a full public sale. That experience has shaped how banks are now approaching the Anthropic-linked Texas project, applying lessons learned from a market still absorbing an unprecedented volume of AI debt. Investors evaluating AI infrastructure debt continue to weigh construction risk alongside the financial strength and credit profile of major tenants and project backers. Anthropic’s long-term lease and Google’s financial backing provide credit characteristics that distinguish this financing structure from projects without comparable contractual support.

Google’s involvement gives the project a credibility boost, but the backstop only takes effect once construction crews finish the data center and operators place it into service. Until then, bond investors carry the risk themselves, including the possibility of construction delays, cost overruns, and other setbacks common to massive infrastructure builds. That gap between construction risk and completed-asset security explains why rating agencies are expected to assign the bond a speculative-grade rating despite Google’s backing. Investors buying into this deal are effectively betting on execution as much as on the underlying tenant relationship with Anthropic. The structure reflects a project finance approach in which long-term contractual support begins after completion, while lenders and investors shoulder most of the construction risk during the build phase. Buyers who accept that risk typically demand higher yields to compensate for the uncertainty involved.

Power, Water and the Cost of Building in Texas

The Hubbard campus will run on its own natural gas power plant, a choice that helps developers sidestep the delays and rising costs of connecting to Texas’s overburdened electrical grid. Texas has become one of the most active states for data center construction, and that surge has raised real concerns about strained power and water supplies across the region. The rapid expansion of data centers has prompted growing scrutiny from utilities, regulators, and local communities over electricity demand, infrastructure investment, and long-term grid capacity. Building a dedicated power plant alongside a data center sounds efficient on paper, but it complicates financing considerably. Lenders now have to underwrite two very different kinds of risk at once: the data center’s technology and operations, and the power plant’s construction and fuel supply chain.

Wall Street has already seen how tricky this kind of combined financing can get. Project Walleye, a comparable Meta-linked data center financing that also includes behind-the-meter power generation, has become a recent reference point for investors evaluating similarly structured AI infrastructure transactions. That precedent gives bankers a useful benchmark as they price the Anthropic-linked Texas bonds before bringing them to market. Market participants will likely evaluate the Hubbard financing alongside other recent large AI infrastructure transactions, including Project Walleye, as they assess project risk. Meanwhile, more AI infrastructure developers are incorporating dedicated power generation into new projects as they seek alternatives to lengthy grid interconnection timelines. Developers increasingly view on-site power generation as a necessary trade-off against grid delays, even though it raises the financing bar considerably.

Chips, Custom Silicon and Separate Financing Streams

The data center itself will house Google’s custom TPU chips, and a separate financing arrangement will cover that hardware instead of the $15 billion construction package. Keeping chip financing distinct from the building and power infrastructure allows each piece of the deal to carry its own risk profile. TPUs depreciate and follow a different upgrade cycle than physical buildings and power plants, so lenders generally prefer not to bundle those risks together. Separating equipment financing from real estate and infrastructure financing follows established project finance principles, allowing lenders to structure each asset class according to its own risk profile. It reflects a maturing playbook where developers finance each layer of an AI data center, from land to power to silicon, on its own terms.

Google’s financial support for the Hubbard project first came to light in March, and separate reporting revealed additional financing details last week. Nexus and Morgan Stanley have both declined to comment on the record, while Anthropic and Google did not respond to requests for comment on the arrangement. Although the companies have not commented publicly, the project’s financing structure combines a long-term Anthropic lease with Google’s financial backing, features that lenders and market participants consider central to the transaction. This deal offers a clear signal of where AI infrastructure financing is headed: bigger packages, faster turnover from banks to bond markets, and increasingly complex risk-sharing structures between developers, lenders, and tech giants. In turn, the bond market’s growing appetite for this kind of debt could reshape how developers build and fund future AI campuses across the country.