OpenAI sidesteps Nvidia with unusually fast coding model on plate-sized chips
Summary
Ars Technica reports OpenAI's GPT-5.3 Codex Spark runs on Cerebras plate-sized chips, delivering coding at about 1000 tokens per second (roughly 15x faster than prior Codex). The move away from Nvidia highlights a broader hardware diversification and a race among AI coding agents to maximize speed for rapid prototyping, with tradeoffs in depth and potential accuracy, and significant industry shifts toward alternative chipmakers and cloud partners.