Anthropic expands AI compute deal with Google, Broadcom

Anthropic expands AI compute capacity with Google and Broadcom, boosting TPU use after demand surged; new resources are expected in 2027.

Jason Kwon ·

Anthropic expands AI compute deal with Google, Broadcom

Anthropic said on Monday it has expanded an agreement with Google and Broadcom to raise processing and compute capacity for its Claude AI models, citing a sharp increase in demand. The company described the move as a major infrastructure step intended to keep pace with rapid growth in usage of its models.

Under the updated arrangement, Anthropic will increase its use of Google Cloud’s Tensor Processing Units (TPUs), which it described as advanced chips designed for AI workloads. The company said the additional compute is meant to support continued development of Claude and to handle a growing customer base.

The announcement builds on an earlier deal reached in October 2025 that covered more than one gigawatt of compute capacity. Anthropic said the newly added compute resources are expected to come online in 2027, indicating a multi-year buildout timeline rather than an immediate capacity increase.

Anthropic did not provide the precise size of the expansion in its statement. However, a recent Broadcom filing with the U.S. Securities and Exchange Commission indicated the agreement covers 3.5 gigawatts of compute, offering a figure for the scale referenced in the announcement.

The company said most of the capacity will be located in the United States. That footprint aligns with Anthropic’s previously stated commitment to invest $50 billion in U.S. compute infrastructure, positioning the expanded buildout as part of a broader domestic infrastructure plan tied to its AI roadmap.

For global markets, the announcement underscores how leading AI developers are seeking long-duration access to specialized compute, often through partnerships that combine cloud platforms and chip supply chains. The involvement of Google Cloud and Broadcom highlights the role of hyperscale cloud infrastructure and semiconductor providers in meeting demand for large-scale model training and deployment.

Key details remain unclear from Anthropic’s announcement, including the exact incremental capacity beyond the October 2025 agreement and how the expanded resources will be phased in before 2027. Anthropic’s statement focused on the need for a significant upgrade, while the SEC filing provides a capacity figure but does not resolve the full timeline or deployment specifics.

More stories