EP 610 : OpenAI's Jalapeño Chip Beats Nvidia GPUs – Power, Speed & New AI Silicon Race

🔥 This week’s AI roundup turns up the heat with OpenAI’s brand‑new in‑house chip, Jalapeño. In its first benchmark release the 700‑watt silicon outpaces Nvidia’s flagship GPUs, delivering up to 3.6× faster inference and 1.9× more work per watt. Built in just nine months with help from OpenAI’s upcoming Astra model and Codex, Jalapeño is designed for inference‑only workloads and will stay inside OpenAI’s data centers, not on the open market. Two follow‑up generations are already in the pipeline, promising even tighter token costs and faster turn‑around for future models.💡 Across the lab‑silicon frontier, Caltech professor Anima Anandkumar and engineer Benedikt Jenik launched Accelerated Understanding, a startup that swaps classic transformers for “neural operators” to predict how the physical world evolves. Their prototype processes 5 trillion data points in a single run—millions of times more than today’s top Google or Anthropic models—and is targeting chip‑materials, extreme‑weather forecasting and robotics.🖥️ Perplexity teams up with Nvidia to ship Portable Computer, a one‑click install that runs the Perplexity “Computer” agent locally on Nvidia’s $4,699 DGX Spark desktop. Users can pick Qwen 3.8 27B or Perplexity’s own 27B model, keep data private, and fall back to 15+ cloud models when needed—all without spending credits.⚡ In other headlines: ChatGPT Work now offers password‑filled browser automation; Anthropic’s Claude gains cross‑tool memory; Google rolls out industry‑specific agents for legal and finance; Apple debuts an $899 Mac Mini with a new M6 chip; and Anthropic hints at a $30 trillion addressable market as it eyes an IPO.🔧 Today’s top AI tools include AdKit, Skydive, Sparkvid, Adapt, Portable Computer, Keenable, and the new Stickers in ChatGPT. Dive in, share the episode, and hit subscribe for the next wave of AI breakthroughs.Tools mentioned: AdKit, Skydive, Sparkvid, Adapt, Portable Computer, Keenable, Stickers in ChatGPT.