OpenAI's recent chip deals heap more pressure on TSMC

OpenAI has secured significant deals with AMD and Broadcom for AI chip production. These agreements involve vast quantities of GPUs and AI accelerators, with initial deployments expected in late 2026. The scale of these deals necessitates hundreds of billions of dollars. Experts suggest Broadcom will assemble existing components for OpenAI's specific needs, particularly for inference tasks. The gigawatt figures reflect uncertainty in exact chip quantities required. These custom chips aim to reduce costs and improve performance for OpenAI compared to buying from NVIDIA. A key benefit is supply chain diversification, avoiding reliance on a single vendor. However, all advanced AI chips, regardless of design, are manufactured by TSMC. TSMC's dominance makes it a critical single point of failure for the global economy. The company's advanced manufacturing processes, particularly EUV lithography, have propelled its success. Competitors like Intel and Samsung are struggling to match TSMC's yield and customer base. TSMC's high profit margins stem from its mastery of chip fabrication. This technical expertise creates a strong lock-in for companies designing chips for TSMC. The reliance on TSMC creates a manufacturing bottleneck, with constrained supply expected to persist. This capacity shortage has already impacted companies like NVIDIA. TSMC is investing heavily in expanding production, including facilities in the US. Nevertheless, global reliance on Taiwan for cutting-edge chip manufacturing will likely continue for years.