Anthropic is negotiating direct leases for multiple hyperscale data center facilities to support its AI models, leveraging Google’s infrastructure services. Meanwhile, Google’s TPU sales are seeing orders redistributed among MediaTek, Marvell and Broadcom, indicating intensifying competition in the AI chip market.
Anthropic has initiated talks with major operators to secure direct leases for multiple hyperscale data center facilities to handle growing AI training and inference workloads, building on its existing integration with Google’s network and power infrastructure. By leasing capacity directly, Anthropic aims to lock in long-term rates and optimize facility configurations for its advanced models.
Google’s custom Tensor Processing Unit (TPU) orders are now being split among MediaTek, Marvell and Broadcom rather than fulfilled solely by Google, reflecting a strategic redistribution of AI accelerator production. MediaTek and Marvell have secured sizeable initial allocations, while Broadcom has obtained a smaller share, underscoring intensifying competition for AI chip supply.
These moves could boost Google Cloud’s infrastructure revenue through higher data center utilization and chip licensing fees, but competition from third-party suppliers and long-term lease commitments may compress margins. Market participants will monitor how these developments affect Google’s hardware costs, cloud profitability and positioning against rival cloud and AI infrastructure providers.
