Google began developing tensor processing units in 2013, initially to make services such as Search more efficient. For years, those custom chips were largely an internal advantage: customers buying Google’s AI offerings were buying into a tightly controlled stack of chips, models and cloud infrastructure.¹

That playbook began to change in the second quarter, when Google opened TPUs to outside data centers. On Tuesday, Alphabet and Oracle went further, announcing that Oracle cloud customers will be able to use Google TPUs from next year — the first time the chips will be available through another hyperscaler’s platform.²

The shift turns Google’s internal silicon program into a more direct commercial challenge to Nvidia, whose GPUs still dwarf every rival’s chip business. Nvidia chief executive Jensen Huang had earlier dismissed the TPU threat, asking: “Without Anthropic, why would there be any TPU growth at all?”¹ But Google’s customer list is widening. Anthropic has committed to vast Google compute spending while also using Amazon, AMD and Nvidia hardware; OpenAI is reportedly preparing to buy TPUs as well. The emerging picture is not a wholesale Nvidia retreat, but a multichip strategy designed to give AI companies more leverage on price, performance and supply.¹

Google argues it has a cost advantage precisely because it controls the full stack. Cloud chief Thomas Kurian said its AI servers pay back in under two years, while systems based on Google silicon do so in roughly half that time: “It’s because we can optimize that whole stack.”²

Yet the Oracle deal exposes the difficult part of the expansion. Selling chips outside Google can add customers and revenue, but analysts estimate TPU sales carry thinner margins than Google Cloud overall. Kurian must also ensure that demand from external buyers does not crowd out Google’s own AI ambitions. CFO Anat Ashkenazi has signaled the bigger TPU revenue payoff is expected in 2027 — a timetable that gives Google room to scale, but gives Nvidia time to defend its lead.²