Wall Street Tightens AI Lending Even With Google Guarantees

Some U.S. banks are refusing new AI-related loans even when the deals carry guarantees from Google, according to a Reuters report published on November 10, 2026. The refusal marks a sharp reversal for a sector that, until months ago, absorbed capital at record pace, and it lands directly on Anthropic, which filed for an initial public offering the same day targeting up to USD 64.4 billion.

The numbers define the squeeze. Alphabet's third-quarter filing shows USD 75.7 billion in credit support commitments backing data center rent and certain electricity costs as of the end of September, of which USD 48 billion covers contracts already signed with third-party data center operators. Moody's calculated in February that major large cloud companies carry USD 969 billion in future rent commitments, with USD 662 billion absent from balance sheets because the leases have not yet commenced. Banks now want to know who pays when the collateral is a chip.

What exactly has changed on Wall Street?

According to Reuters, major U.S. banks and credit investors have tightened scrutiny of AI-related loans and turned cautious on additional funding, with some banks reluctant to extend new loans even for transactions backed by guarantees from Anthropic or Google. The review process now includes stress-testing cash flows under recession scenarios and examining data center commissioning deadlines, service obligations, and the conditions under which major customers can reduce or terminate contracts.

In plain terms, lenders no longer take contracted rent and computing-service revenue at face value. They verify it.

Collateral valuations show the same conservatism. Some lenders demanded expanded guarantees for Nvidia's USD 500 billion AI financing initiative, arguing the long-term collateral value of its chips is difficult to assess. Loans secured by graphics processing units (GPUs, the accelerators that train and run AI models) are typically underwritten on a 3–4 year depreciation schedule, because data on profitability and resale prices beyond that window is insufficient.

Customer creditworthiness has become the decisive factor. The investment-grade rating on CoreWeave's USD 8.5 billion GPU-backed loan rested heavily on the stability of payments Meta is contractually obligated to make, illustrating that lenders now weigh the obligor as much as the asset.

Banks are also shedding existing exposure. Per the Financial Times, major banks spent more than six months distributing USD 38 billion in construction loans for data center projects leased by Oracle in Texas and Wisconsin to outside investors, with some banks selling the loans at a discount. Others are weighing arrangements in which they retain the loans but pay private credit funds and insurers to assume part of the loss risk. Concentrated lending to the AI sector has reduced their capacity to fund anything new.

Why does the Google–Anthropic guarantee matter?

On November 4, AI data center infrastructure company TeraWulf disclosed an agreement under which Google provides credit enhancement for Anthropic's 20-year, USD 19 billion lease of a data center in Kentucky. If Anthropic fails to pay rent or becomes insolvent, Google would pay the outstanding rent and take over as tenant, or make a termination payment under certain conditions. The developer intends to use that backing to secure construction financing.

Some banks still refuse to lend against it. Reuters reports this reluctance reflects efforts to manage total lending exposure to the AI sector, including the companies providing the guarantees — a signal that Big Tech credit is no longer an automatic pass.

The counterparty risk concentrates in one place. Because rent payments across multiple projects depend on the revenue of a small number of AI companies, guarantees extended separately across individual contracts carry significant risk of becoming payable at the same time.

Anthropic itself illustrates the tension. The company recorded an operating loss of USD 567 million in the third quarter, driven by computing costs, even as revenue rose 47% from the previous quarter. Continued losses, combined with rising long-term spending commitments, fuel concerns that debt-and-guarantee-financed expansion could end in overinvestment. If utilization of computing resources secured through long-term contracts falls short, reliance on external funding rises further. Anthropic's own securities filing identifies the risk that constraints on infrastructure partners' external financing or credit support could delay or reduce the supply of contracted computing resources or trigger price adjustments.

How large are Big Tech's hidden obligations?

Per the Financial Times, residual value guarantees that technology companies provided or committed through financing programs over the preceding year total up to USD 300 billion. Under these structures, a special-purpose vehicle (SPV, a standalone entity created to hold specific assets) borrows to acquire data centers and semiconductors, while the tech company covers any difference between the assets' disposal price and an agreed minimum value.

The structure has spread. The guarantee model used for Meta's data centers now also backs financing for AI chips supplied by Broadcom to Anthropic, supporting customers' facility expansion without recording all related borrowings as direct debt.

Moody's warned in February that renewal periods and residual value guarantees in data center leases are not adequately reflected in reported lease liabilities. When renewal is not judged reasonably certain, extension-period rent is excluded from liability calculations. The consequence: book debt alone cannot reveal long-term payment obligations, and as facilities enter operation in stages, deteriorating AI profitability could pressure both investment capacity and cash flow simultaneously.

How are falling tech stocks compounding the problem?

Alphabet has been raising money at scale. In June, it completed a public common stock offering, a private placement to Berkshire Hathaway affiliates, and an issuance of mandatory convertible preferred stock, recording net proceeds of USD 20.5 billion, USD 10 billion and USD 19 billion respectively in its third-quarter filing, with AI infrastructure capital spending listed among the uses. It also established a program to sell up to USD 40 billion in shares into the market over time, raising USD 12.4 billion by the end of September.

Share prices have turned against that machinery. Alphabet stock fell roughly 19% during October and dropped more than 12% in a single day on October 29, immediately after its earnings announcement. Raising the same amount at a lower price requires issuing more shares, increasing dilution. The at-the-market program depends on prices at the moment of sale, so a prolonged downturn makes the size and timing of additional issuance harder to plan.

What happens to Anthropic's IPO price?

In a securities filing submitted on November 10, Anthropic set a target offering price of USD 70–78 per share and said it would offer 825 million shares, including shares sold by existing shareholders. At the top of the range, the offering totals USD 64.4 billion, with a fully diluted valuation of USD 1.4 trillion including stock options and other instruments.

That is below the USD 2 trillion some investors previously hoped for, and the wariness runs deeper. Reuters reports some prospective investors questioned even a previously discussed USD 1.5 trillion valuation, citing enormous infrastructure spending commitments and volatility in AI-related stocks. Potential investors have discussed valuations around USD 1.2–1.3 trillion, with some considering figures as low as USD 1.1 trillion.

The arithmetic is unforgiving. Based on the proposed offering volume, every USD 1 reduction in the per-share price cuts total offering proceeds by USD 825 million. With banks tightening lending, an underperforming listing would force Anthropic back to the market to fund infrastructure commitments it has already signed — the same market that just closed the door on Google-guaranteed loans.