OpenAI's first official Jalapeño benchmarks: up to 1.9x more work per watt and 3.6x lower latency than the best recorded GB200/GB300 results — the numbers, context, and honest limits.
AMD acquires Canadian startup Taalas for technology that etches AI model weights directly into silicon, achieving 48x NVIDIA inference speed. Full analysis of the technology, numbers, and impact.
Meta will begin manufacturing its new AI chip codenamed Iris in September 2026, joining the MTIA family to cut reliance on Nvidia and AMD. Full specs, comparisons, and market implications inside.
Start typing to search
