OpenAI's first AI chip brings the heat
OpenAI unveiled Jalapeño, its first custom AI chip built with Broadcom, which benchmarks show outperforms Nvidia's flagship GPUs on speed and power efficiency. The newsletter also covers emerging physical AI startups, new AI agent tools from Perplexity and Claude, and a community member's successful church construction project management app built with AI.
Summary
OpenAI released benchmark results for Jalapeño, a custom-designed chip optimized for running (not training) AI models. The 700-watt chip demonstrated superior performance compared to Nvidia's 1,200-watt systems, achieving up to 3.6x faster response times and 1.9x more work per watt. OpenAI developed the chip in nine months using its Astra model and Codex, though the company won't commercialize it due to high internal demand. The company continues relying on Nvidia for model training while planning two additional chip generations to deploy across its data centers throughout 2027. This trend reflects broader industry movement, with Google, Anthropic, Amazon, and Microsoft all developing proprietary silicon.
Accelerated Understanding, a new startup founded by Caltech professor Anima Anandkumar and engineer Benedikt Jenik, is pursuing physics-based AI rather than traditional transformer architectures. Their neural operator-based model processed 5 trillion data points in a single run—approximately 5 million times more than Google and Anthropic's leading models—and targets applications in chip materials, extreme-weather prediction, and robotics. The founders notably declined a 35% stake and executive positions at Jeff Bezos' Prometheus to build independently, underscoring their conviction in the physics-based approach.
Perplexity and Nvidia launched Portable Computer, an on-device version of Perplexity's Computer agent that runs locally on Nvidia's $4,699 DGX Spark hardware, offering privacy benefits and free local computation. Users can select between open models or access cloud alternatives, with the option to integrate with 15+ cloud services. Claude introduced a shared memory feature across its chat and Cowork products that updates in real-time, while Apple positioned its new $899 Mac Mini as purpose-built for 'always-on agentic computing' with M6 chips enabling 4x faster workload processing.
The newsletter featured a community workflow from a 71-year-old retired cattle rancher building a construction management application in Replit for his church's $13.5M renovation project, demonstrating how non-programmers can use AI to bridge technical and domain expertise gaps.
About this episode
PLUS: Build a website hands-free with Claude Voice
Key Insights
- OpenAI's internally-designed Jalapeño chip achieves 3.6x faster performance and 1.9x better power efficiency than Nvidia's flagship systems, though OpenAI will not commercialize it due to internal demand exceeding supply.
- Accelerated Understanding's physics-based neural operator model processed 5 trillion data points in a single run, approximately 5 million times more than Google and Anthropic's transformer-based models, suggesting a fundamental architectural advantage for physical world prediction.
- Accelerated Understanding's founders declined a 35% equity stake and executive positions from Jeff Bezos' $12B-funded Prometheus venture to pursue their own company, demonstrating confidence in their physics-over-text approach to AI.
- Perplexity's shift toward on-device inference through Portable Computer reflects industry movement to localize AI computation for privacy and cost reduction, with the added benefit that smaller open models can serve as primary systems rather than backups.
- Non-technical domain experts can now build production AI applications by describing requirements in plain English and iteratively refining AI-generated code, as demonstrated by a 71-year-old retiree creating a complex construction management system without programming experience.
Topics
Transcript
Good morning, {{ first_name | AI enthusiasts }}, and welcome to our 3,221 new readers. OpenAI named its first in-house AI chip Jalapeño, and the first benchmarks just came out appropriately spicy. The company's internal tests have the chip beating Nvidia's flagship GPUs on both speed and power efficiency, with a design OpenAI credits its Astra model and Codex for helping create in nine months. OpenAI's Jalapeño chip brings the heat New startup takes AI from words to physics Build a website hands-free with Claude Voice Perplexity, Nvidia go portable with Computer OPENAI The Rundown: OAI just published the first benchmark results for Jalapeño, the custom chip it built with Broadcom to run AI models (not train them), with its…
Full transcript available for MurmurCast members
Sign Up to AccessMore from The Rundown AI
OpenAI’s “generational leap” with GPT-6 Astra
OpenAI released GPT-6 Astra, positioning it as a major advancement in AI with exceptional benchmark performance across multiple domains. The newsletter also covers Google's improved weather forecasting model, the Loop Method for ChatGPT optimization, and a reader's positive-news-only AI app.
Meta, Google join the AI launch party
Meta and Google launched new AI models in early September, with Meta's Muse Spark 1.3 achieving near-frontier performance at low cost while Google's Gemini 3.8 Flash represents a recovery step but still trails the frontier. The newsletter also covers AI safety concerns about reasoning transparency, tech literacy as a career ceiling, and practical AI workflows for professional development.
Fable 5.1 kicks off launch week at the frontier
Anthropic released Claude Fable 5.1, showing significant improvements in coding and research tasks with reduced safety rejections, marking the end of the summer's cautious release period as OpenAI's Astra launch approaches. Bernie Sanders published an op-ed calling for a global AI pause, citing control concerns and societal risks, while ongoing legal battles between Apple and OpenAI center on alleged theft of confidential designs.
Runway's Solaris previews the no-code internet
Runway unveiled Solaris, an AI-powered interface that renders websites and apps as real-time video with no underlying code, while Imperial College researchers developed an AI model that detects heart disease from ECGs in under two seconds with superior accuracy to human doctors.
OpenAI cuts out SpaceX-owned Cursor
OpenAI is removing its models from Cursor coding editor by mid-November following SpaceX's acquisition of the platform, citing Elon Musk's history of contract violations as justification. The move escalates the ongoing feud between Sam Altman and Elon Musk while putting developers in the middle of a high-profile corporate conflict.