OpenAI is building internal software to run AI workloads across hardware from multiple manufacturers, reducing reliance on Nvidia.

The Stargate data center scaling initiative provides the broader context and scale for these diversification efforts.

In the same period, OpenAI struck a deal with AMD to deploy up to 6 gigawatts of AMD Instinct GPUs, with deployment slated to begin in the latter half of 2026.

OpenAI formed a Broadcom partnership in late 2025 to co-develop accelerator and networking systems, signaling a push toward diversified hardware ecosystems.

Current job postings show OpenAI hiring to build automatable, repeatable systems for heterogeneous clusters, underscoring practical implementation of this multi-vendor approach.

OpenAI’s multi-vendor strategy parallels moves by Google with TPUs, Amazon with Trainium and Inferentia, and Microsoft with Maia.

OpenAI’s Hardware group is designing its own silicon, collaborating with TSMC on chiplet architectures and advanced packaging, with custom inference chips expected by 2026.

If successful, in-house inference chips and diversified hardware could reduce per-query costs and alter the economics of AI services for consumers.

Taken together, these developments could reshape the AI infrastructure landscape by reducing Nvidia’s near-monopoly and building alternative ecosystems for AI workloads.