{ "tldr": { "summary": "Nvidia reinforced its AI chip dominance by topping the MLPerf Inference v6.0 benchmarks with record performance, driven by software optimizations that cut costs and boost throughput. Google's continued absence from these benchmarks undermines its competitive positioning in the AI hardware market.", "key_points": [ "Nvidia set new performance throughput records across AI models in MLPerf Inference v6.0, including for the reasoning model DeepSeek-R1.", "Software optimizations increased Blackwell Ultra token throughput by up to 2.7x over six months, reducing cost per token by over 60%.", "Nvidia's system-wide codesign and software improvements are key to its performance gains, making inference more economical for deployments.", "Demand for Nvidia's Blackwell Ultra systems is outpacing manufacturing capacity due to continuous performance improvements.", "Google did not submit its TPU v7 Ironwood for MLPerf v6.0, extending a pattern of non-participation in benchmarks.", "Google's benchmark no-shows may hurt its ability to win AI chip customers and challenge Nvidia's market leadership." ] }, "trade_ideas": [] }