Skip to content
Artificial intelligence. Clearly. RSS feed
AI NEWS AI

News. Evidence. Perspective.

BusinessAnalysis

What Google gets for backing a data-centre tenant’s rent

A payment promise can help a developer borrow and give Google an equity opportunity. The contracts and Alphabet’s accounts show the obligations, choices and potential recoveries behind that exchange.

Google received warrants to buy more than 41 million TeraWulf shares at one cent each under an August 2025 agreement. Its contribution to that transaction was a promise: stand behind specified rent obligations of Fluidstack, the tenant taking space in two AI computing buildings at Lake Mariner in New York.[1]

TeraWulf described the initial backstop as $1.8 billion supporting project debt financing. The buildings were contracted for more than 200 megawatts of critical IT load, with ten-year leases. This gives the landlord something useful before the computing business has generated years of cash: another company with substantial financial resources can become responsible if the tenant fails under the agreed conditions.[2]

The economic exchange has three sides. The developer gains a more financeable rental contract. The tenant gains access to infrastructure. Google receives a potential ownership interest in the developer while helping capacity get built. The one-cent figure is the warrant exercise price; the consideration for receiving the warrants is the payment exposure Google accepts. Existing shareholders can be diluted when those warrants are exercised.[1]

What the promise requires

The backstop starts when the corresponding lease commences. Following a qualifying payment default or insolvency event and notice, Google can be required to pay outstanding rent and take over as tenant. The original filing also describes a route in which it pays a termination fee. This supports specified lease obligations; it is not a blanket promise to repay every debt TeraWulf might incur.[1]

The actual recognition agreement makes the choice narrower than a quick summary suggests. The termination route is available if the trigger occurs after the first six years, or if installed equipment is capable of consuming less than half the baseline capacity. Otherwise, Google must assume the lease when an election is required. A payment default without insolvency also permits a deferral while Google pays unpaid and ongoing rent. If it pays the termination fee, it receives an equity interest in the project company calculated from that payment and the project’s development and construction cost, including financing costs.[3]

That matters to the eventual economics. A payment can leave Google with a lease it can use, or a project interest, rather than simply an unrecoverable bill. The value of those rights depends on the usefulness and value of the infrastructure when they are needed. It does not follow that the recovered asset will be worth the cash paid.[3][4]

Two numbers describe different things

Alphabet’s June 2026 accounts report its wider data-centre payment backstops as credit derivatives. They distinguish the maximum potential exposure under specified defaults from the fair-value liability recognised on the balance sheet.[4]

In detail

Alphabet’s data-centre credit derivatives at 30 June 2026

Alphabet’s data-centre credit derivatives at 30 June 2026
MeasureUS dollars
Gross notional exposure$43.785 billionSource 4
Recognised fair-value liability$815 millionSource 4

Alphabet says valuation uses probability-weighted models incorporating counterparty risk and contractual terms. Its potential exposure falls as counterparties make payments; some termination payments can be offset by equity or cash receipts that are excluded from the gross notional figure. The two accounting numbers therefore answer different questions, and subtracting one from the other does not reveal a hidden loss.[4]

Credit protection has a provider

S&P Global Ratings explicitly identifies Google’s Fluidstack backstop as protection for creditors. Its March credit FAQ also explains the other side: support for weaker counterparties can enter the guarantor’s adjusted debt, with allowance for factors such as recoverable asset value. A credit analyst can consequently assess more exposure than appears in a company’s reported borrowing alone. That is an analytical adjustment, not an assertion that all guaranteed payments are already due.[5]

For Google, the investment proposition combines potential warrant value, an interest in additional computing infrastructure and the cost of standing ready to pay. A return calculation would need the value ultimately realised from those benefits and any payments or losses along the way. The disclosed exercise price and headline backstop amount cannot supply that answer. What they do reveal is a form of capital that helps the AI build-out proceed: someone willing to make another company’s payment promise more credible.

Follow the evidence

Sources & further reading

  1. Credit FAQ: How Could AI Risks Weaken Ratings? S&P Global Ratings · 11 Mar 2026 · Accessed 19 Oct 2026