Broadcom

Overview

Broadcom appears in the sources as a specialist implementation and connectivity partner for custom AI accelerators. It contributes the physical, mixed-signal, network-topology, and back-end capabilities that turn a model company’s architecture into a deployable chip and multi-chip system.

Current Profile

The current profile spans two custom-silicon relationships. In Google’s TPU system, Broadcom is tied to ICI, physical links, topology details, and back-end implementation. In E251, it is named as OpenAI’s partner on HanaPino. Both cases show that custom accelerator strategy depends on implementation and system integration, while also creating supplier bargaining and execution dependencies.

Key Characteristics

• Provides back-end and physical implementation capability for custom AI chips.
• Contributes mixed-signal, connectivity, and topology expertise to multi-chip systems.
• Helps model and cloud companies convert workload requirements into deployable silicon.
• Acts as both a capability multiplier and a strategic supplier dependency.

Evidence

• TPU implementation: E228 ties Broadcom to ICI, back-end work, physical chip links, and topology-network implementation for Google TPU Pods.
• OpenAI custom chip: E251 identifies Broadcom as the partner behind HanaPino and discusses the chip’s HBM and power-oriented design.

Qualifications

The sources describe Broadcom’s role at a high level and do not specify contractual boundaries, IP ownership, economics, exact division of labor, or independent performance attribution. HanaPino details remain source-scoped.

What Changed

• Migrated the page to the synthesis-v1 entity schema.
• Extended Broadcom from the Google TPU relationship to OpenAI’s source-reported HanaPino program.
• Reframed implementation expertise and supplier dependence as two sides of the same profile.

Relationships

• Google - customer and system owner in the TPU implementation case.
• TPU - accelerator family where Broadcom supports physical and system implementation.
• OpenAI - model company named as Broadcom’s HanaPino partner.
• HanaPino - custom inference chip linked to Broadcom in E251.
• AI Cluster Networking - connectivity layer where Broadcom’s mixed-signal and topology expertise matters.
• Strategic AI Infrastructure Dependence - dependency and bargaining frame created by specialist implementation partners.