OpenAI’s GPT-6.1 Sol Ultrafast Uses NVIDIA Chips, Not Cerebras

30 Sept 2026, 1:12 pm · 2d ago · 1 min read · OfficeChai

OpenAI's newly announced Ultrafast tier for GPT-6.1 Sol is powered by NVIDIA GPUs rather than Cerebras wafer-scale hardware, according to an analysis by research firm SemiAnalysis. Despite OpenAI's ongoing strategic partnership and exploration of alternative AI hardware accelerators for faster inference workloads, NVIDIA remains the primary infrastructure backbone for powering its latest low-latency enterprise model tiers. The revelation highlights the sustained dominance of NVIDIA’s hardware and CUDA software ecosystem in production-grade AI inference, despite growing industry efforts to diversify hardware suppliers and reduce dependency on the leading chipmaker for high-speed computing.