Filing: Broadcom agrees to produce future versions of Google's TPUs and expands its Anthropic deal to give the startup access to ~3.5 GW of computing capacity
- Broadcom said it agreed to produce future versions of Google's artificial intelligence chips,
Context & Ripple Effects
This extends an already material TPU commitment: Anthropic had previously placed large orders for Google's Ironwood TPU racks through Broadcom. The new arrangement ties together Google's chip roadmap, Broadcom's manufacturing role, and Anthropic's access to multiple gigawatts of next-generation capacity rather than treating the startup as a spot buyer of compute.
It also resolves, for now, an earlier supply-chain question: Google was reported to be considering MediaTek for a next-generation TPU while retaining Broadcom, but the filing makes Broadcom's role in future TPU production explicit. Anthropic's separately reported revenue acceleration gives the capacity commitment a clearer commercial rationale.
First-order effects
- Broadcom gains a continuing production mandate for future Google TPUs, strengthening its position in Google's custom-AI-chip supply chain.
- Anthropic obtains access to roughly 3.5 GW of compute capacity under an expanded Broadcom arrangement, giving it a defined path to scale workloads on Google's TPU platform.
Second-order effects
- The deal makes TPU capacity a more consequential alternative for frontier-model builders, increasing pressure on competing infrastructure providers to offer comparable scale, availability, and commercial terms.
- Google, Broadcom, and Anthropic become more operationally interdependent: Anthropic's planned capacity depends on TPU delivery and deployment, while Broadcom's manufacturing commitment is more directly connected to demand from a major AI customer.
Third-order effects
- If repeated across major model developers, long-duration compute agreements could shift AI infrastructure from short-term cloud procurement toward utility-like capacity planning, where chip roadmaps and power-scale deployment are negotiated years ahead.
- The arrangement reinforces a hardware-strategy split: leading AI companies may secure differentiated capacity through vertically coordinated custom silicon stacks rather than relying on a single general-purpose accelerator ecosystem.
The trend: This is one data point in the platformization of AI infrastructure, where custom-chip suppliers, cloud platforms, and model developers lock in large, durable capacity relationships.