Google Cloud CEO: Custom AI accelerator payback half that of GPU systems

At the Goldman Sachs Communacopia + Technology Conference in September 2026, Google Cloud CEO Thomas Kurian said the average payback period for the company's AI servers is less than two years, and its custom chips pay back in half the time of GPU-based solutions.

Kurian said Google Cloud's custom AI accelerator business is more than twice the size of the second-largest hyperscaler, and that most of its infrastructure contracts are five-year commitments.

He added that end-to-end control from chips to models lets Google offer products such as Gemini Flash 3.8 at half or less of rivals' prices. According to Kurian, Google Cloud's AI training, AI inference, and CPU offerings provide cost-performance advantages of 2.7x, 80%, and 30%, respectively.

Google Cloud's enterprise AI platform Gemini Enterprise runs machine learning models at 40% of competitors' cost while delivering higher accuracy, which Kurian cited as one reason more than 90% of Fortune 100 companies use the platform.