Why Circular AI Deals Among OpenAI, Nvidia, AMD Are Raising Eyebrows (www.bloomberg.com)

Reports have surfaced describing a web of reciprocal commercial arrangements connecting major AI players — notably chipmakers Nvidia and AMD and leading model developers such as OpenAI — that critics label “circular” because money, hardware and preferential access flow in multiple directions. The concern is not just who pays whom but how those deals shape real-world access to AI compute: preferential inventory allocation, custom firmware and software optimizations for specific accelerators, and bundled support can give favored developers faster training cycles and lower costs while locking out smaller labs and cloud competitors. For the AI/ML community this matters technically and economically. On the technical side, model performance and reproducibility are increasingly tied to specific accelerator architectures (e.g., proprietary drivers, libraries and kernel optimizations), creating hardware lock‑in and fragmentation between CUDA-optimized and ROCm/other stacks. Economically, concentrated supply agreements can raise prices, reduce competition for GPUs, and skew research priorities toward models that exploit vendor-specific features. The net effect could slow innovation, limit independent benchmarking, and draw regulatory scrutiny; remedies likely to be discussed include greater transparency in procurement, open standards for runtimes and kernels, and diversified supply chains to preserve fair access to large-scale compute.