Google extends Intel SmartNIC partnership as custom chip business tops $1bn on run rate

Google will continue purchasing SmartNICs from Intel for its public cloud infrastructure and the two companies are jointly developing next-generation infrastructure processing units, Intel said, in a deal that underlines the growing commercial weight of custom silicon for hyperscale data centres.

SmartNICs, also referred to as infrastructure processing units (IPUs), are network cards equipped with onboard processors that offload networking, security and storage tasks from a server's main CPU, freeing that capacity for the workloads run by cloud tenants.

Intel and Google previously collaborated on Mount Evans, an ASIC-based IPU, short for application-specific integrated circuit, which shipped alongside Google Cloud's C3 compute instances and represented the first generation of their joint custom silicon effort.

Intel did not disclose technical specifications or a deployment timeline for the next-generation IPUs under development.

The announcement came alongside a disclosure from Intel chief financial officer David Zinsner that the company's custom ASIC business has grown to an annualised revenue run rate above $1 billion, reflecting rising demand from hyperscalers building bespoke silicon to optimise their infrastructure costs and performance.

Google Cloud instances already make broad use of Intel Xeon processors for general-purpose and AI workloads, a relationship Intel said would continue alongside the IPU collaboration.

Google also runs its own Arm-based Axion CPUs across parts of its cloud infrastructure, meaning Intel occupies one of several positions in Google's diversified processor strategy rather than serving as a sole supplier.

Intel and Google provided no further detail on performance targets for the new IPUs or their planned distribution across Google Cloud regions.