Oracle has struck a deal with Google to become the first rival hyperscaler to offer its Tensor Processing Units (TPUs) to cloud customers.
Announced at Oracle's AI World event in Las Vega on Tuesday, the agreement will see Google's TPUs made available through Oracle Cloud Infrastructure (OCI) from 2027.
Neither Oracle nor Google revealed the number of chips involved, nor did they release the financial terms of the pact. They did, however, say that they plan to ramp up to gigawatt scale over multiple years.
In a flurry of press releases, inference chip deals with SambaNova and Cerebras were also disclosed, while Oracle said it is expanding its deployment of Nvidia's Vera Rubin systems to multigigawatt scale too.
