Alphabet is using $43.8B backstops to sell more TPUs

Signals Inbox·August 3, 2026·AI Chips

Alphabet is not only designing TPUs and waiting for customers to buy them. It is putting up to $43.8 billion behind third-party data center obligations, helping developers raise money, build capacity, and fill those sites with Google chips. The real competition with Nvidia is now moving from silicon into balance sheets.

Q1What did Alphabet officially disclose?

In its latest quarterly filing, Alphabet reported $43.8 billion of potential exposure from credit backstops tied to data centers. That was up from $16.9 billion at the end of 2025 and $28.4 billion only three months earlier. It also said another estimated $24.1 billion of future backstops is being prepared.

Q2What is a data center backstop?

Think of Alphabet as a very rich guarantor. A developer builds a data center for a customer that wants Google TPUs. Lenders provide the money because Alphabet promises to step in under certain default scenarios. If the customer cannot keep paying, Google may take over the lease, find another tenant, use the capacity itself, or pay a termination fee.

Q3How does this help Google sell more TPUs?

AI chips are useless without buildings, electricity, cooling and network connections. Those projects cost billions and take years. By reducing the risk for lenders, Alphabet helps third parties build TPU-ready capacity faster without Google paying for every site directly. More financed data centers can mean more places where customers run Google chips instead of Nvidia GPUs.

Q4Is Alphabet really risking the full $43.8 billion?

Not necessarily. The number is the maximum exposure under specific default scenarios, not the amount Alphabet expects to lose. Its recorded liability for these credit derivatives was only $815 million in June. Google may also receive leases, equity, cash or usable data center capacity if it has to step in. Still, the headline number shows how aggressively it is supporting the buildout.

Q5Is Google the only chip company doing this?

No. Nvidia has also used investments, purchase commitments and lease support to help AI infrastructure companies finance new sites. It agreed to buy unused CoreWeave capacity under one deal and has backed major data center projects elsewhere. The difference is that Google is now applying the same playbook to TPUs, turning its balance sheet into another weapon against Nvidia.

Q6What is the real point of tension?

Some of the demand for AI chips is no longer separate from the companies selling them. Chip suppliers are helping finance customers, buildings and power so the ecosystem keeps expanding. That can accelerate real deployment, but it also makes demand harder to read. When a vendor guarantees the infrastructure, strong chip sales do not always mean customers could have financed the expansion alone.

Q7So why does this matter now?

Because the AI chip race is becoming a capital race. Performance still matters, but so does the ability to secure power, unlock private debt and absorb risk for customers. Alphabet can now use one of the world’s strongest balance sheets to push TPUs into more external data centers. Nvidia is no longer competing only with Google’s chips. It is competing with Google’s credit too.