AIGuru

Custom Silicon Race: NVIDIA Vera CPU for AI Agents, OpenAI Jalapeño Inference Chip Benchmarks

Custom Silicon Race: NVIDIA Vera CPU for AI Agents, OpenAI Jalapeño Inference Chip Benchmarks

NVIDIA unveiled Vera CPU at Hot Chips 2026, purpose-built for agentic AI workloads, adopted by SpaceXAI for Grok at massive scale including orbital computing with Starmind (targeted Q4 2027). Vera is positioned as the first CPU designed for AI agents, signaling that agentic workloads are CPU-bound, not just GPU-bound. Separately, OpenAI showed Jalapeño chip benchmarks at Hot Chips, demonstrating strong inference performance vs Blackwell, but deployment not until late 2026/2027. New: @therundownai reports Jalapeño achieves 1.9x perf/watt vs Nvidia GB200/GB300, 3.6x faster responses, designed in 9 months, deploying by year-end. Both developments intensify the custom silicon competition and full-stack co-design trend.

Sources (2)
Updated Aug 26, 2026
Custom Silicon Race: NVIDIA Vera CPU for AI Agents, OpenAI Jalapeño Inference Chip Benchmarks - AIGuru | NBot | nbot.ai