Friday, June 26, 2026
EN·DarkSubscribe
AI Infrastructure · News & Analysis
HomeChips & HardwareReport
Chips & Hardware · Report

Broadcom and OpenAI are partnering on a 10-gigawatt custom AI chip and networking platform, establishing OpenAI as a vertical infrastructure player.

The 10GW scale signals OpenAI's AI workload trajectory and validates custom silicon as necessary to manage AI training/inference costs at hyperscale; Broadcom gains strategic foothold in AI infrastructure stack.
Trade pressSlicast · October 14, 2025 · Global · Source: nextbigfuture.com
importance 85

Broadcom and OpenAI announced a multi-year strategic collaboration to co-develop and deploy 10 gigawatts (GW) of custom AI accelerators and integrated networking systems. This marks OpenAI's entry into designing its own AI processors, building on existing Nvidia and AMD GPU supply deals. The collaboration, valued at over $10 billion, aims to reduce reliance on Nvidia GPUs while leveraging model-specific custom optimizations that can cut energy use by 20-30% compared to generic chips.

OpenAI will embed insights from its frontier AI models—including those powering ChatGPT—directly into custom hardware to enhance performance and efficiency. The custom AI accelerators are optimized for inference tasks, inference-heavy workloads, and large-scale AI training. Broadcom will handle the manufacturing, integration, and deployment of full racks containing these accelerators, paired with its end-to-end portfolio of Ethernet switches, PCIe interconnects, and optical connectivity solutions. This standards-based approach using Ethernet avoids proprietary fabrics like Nvidia's InfiniBand, enabling open, power-efficient AI data center clusters with seamless scale-up within single nodes and scale-out across clusters.

The collaboration has been in development for 18 months with long-standing supply agreements already in place, now formalized through a new term sheet governing rack deployments. Initial deployments begin in the second half of 2026, with full rollout completing by the end of 2029 through phased scaling to meet OpenAI's exploding compute demand. Racks will be deployed across OpenAI's own data centers and those of its partners, including Microsoft Azure.

The deal adds $2-3 billion annually to Broadcom during rollout, reflecting AI's growing importance to the company—now accounting for approximately 25% of Broadcom's revenue, up from 10% before 2023. Broadcom's stock rose approximately 10% on the announcement, adding roughly $150 billion to its market capitalization.

Read the original
Broadcom and OpenAI are partnering on a… · Slicast