Google Breaks From Rental-Only Model as Alphabet Books First Revenue From TPU Sales

Alphabet has quietly changed how it monetizes its in-house artificial intelligence hardware. For years, the company’s tensor processing units (TPUs) were available exclusively through Google Cloud as rented infrastructure. In the second quarter, that model shifted: Alphabet delivered complete TPU systems directly to customer data centers for the first time and began recording revenue from outright hardware sales.

The move marks a notable strategic pivot for a company that has historically kept its custom silicon inside its own cloud footprint. Selling TPU systems opens Alphabet to organizations that operate their own data centers at very large scale but want access to the chips that power much of Google’s internal AI stack.

Early Days, With Revenue Back-Loaded to 2027

The financial contribution so far is modest. Alphabet has signed only a limited set of agreements, targeted at customers running specialized, high-volume AI workloads on their own infrastructure. Importantly, the company indicated that the significant majority of revenue tied to these deals will not land until 2027, meaning the near-term earnings impact will be minimal even as shipments begin.

That timeline suggests the sales program is still in an early ramp phase, with today’s deliveries functioning more as proof points than a material revenue driver.

Market Snapshot

Shares of Alphabet (NASDAQ:GOOGL) traded down 1.3% at $338.46, while Class C shares (NASDAQ:GOOG) fell 1.26% to $335.31 on the day. The company carries a market capitalization of roughly $4.39 trillion and is classified in the Communication Services sector within the Internet Content & Information industry.

By selling rather than only renting its AI accelerators, Alphabet is positioning itself to compete more directly in the broader AI hardware market, where demand from enterprises building out dedicated compute capacity remains intense. The strategy also deepens the company’s involvement in data center hardware, an area it has long managed internally to optimize its own AI training and inference operations.

The program’s long-term scale will depend on how many large infrastructure operators opt for Google’s custom chips over competing accelerators, and on Alphabet’s ability to manufacture and deliver systems at volume.

What to watch

• Alphabet’s next quarterly earnings report, including any updates on the number of TPU system agreements signed.
• Management commentary on the 2027 revenue ramp and whether additional customers are disclosed.
• Details on which customers have received TPU systems and for what workloads.
• Competitive responses from other AI accelerator vendors serving self-managed data centers.

Source: original release