Overview

• Financial Times–sourced reports say OpenAI is close to starting mass production of in‑house AI accelerators, with initial deliveries expected next year.
• Broadcom CEO Hock Tan told analysts a fourth major custom‑chip client committed more than $10 billion in orders, and multiple reports identify the likely customer as OpenAI, though neither company has confirmed it.
• TSMC is reported to handle fabrication using its advanced 3 nm process for the planned accelerators.
• OpenAI is expected to deploy the chips internally to speed model training and inference and to lessen reliance on Nvidia’s GPU ecosystem.
• Key details remain unannounced, including commercial availability outside OpenAI, and analysts note high development costs and execution risks against Nvidia’s entrenched software stack.