Alphabet has granted Anthropic access to 1 million of its tensor processing units and secured a multibillion-dollar pact with Meta Platforms through Google Cloud to run inference workloads. These agreements underscore rising demand for real-time AI processing and validate TPUs as a commercial product beyond internal use.

Discussions with Marvell Technology aim to co-develop two custom AI chips, signaling Alphabet’s intent to diversify its in-house supply chain beyond Broadcom’s TPU portfolio. A broader supplier network could strengthen supply resilience and support accelerated deployment of inference-focused hardware.

By focusing on specialized inference chip designs and leveraging vertical integration, Alphabet is positioning TPUs to challenge Nvidia’s GPU dominance in AI workloads. As token pricing shifts and global AI models evolve, the outcome of these initiatives will be critical for Alphabet’s long-term infrastructure leadership.