OpenAI’s Jalapeño ASIC beats Nvidia on efficiency in AI chip design
OpenAI’s vice president of hardware, Richard Ho, discussed the company’s Jalapeño inference ASIC—developed with Broadcom—in an interview with Tom’s Hardware. The chip, unveiled at Hot Chips 2026, leverages AI within its own design process to outperform Nvidia in efficiency, addressing misconceptions that it’s only optimized for OpenAI models. The interview, a full transcript, explores AI’s role…
Key points
- OpenAI’s Jalapeño ASIC, co-developed with Broadcom, uses AI in its design to surpass Nvidia’s efficiency, per VP Richard Ho’s interview
- Synopsys’ Autopilot platform for AgentEngineer aims for general availability by year-end, automating chip design workflows
- Anthropic warns GLM-5.3’s open-weight model can bypass safeguards, nearing Mythos AI’s hacking capabilities, per report
Tom’s Hardware also highlights Synopsys’ Autopilot platform for AgentEngineer, a new AI-driven chip design tool nearing general availability by year-end. Experts like Domenec Forte (University of Florida) and Simon Moore (University of Cambridge) warn of patent risks as AI reshapes chip design, potentially exposing proprietary intellectual property. Meanwhile, Intel reportedly secured a patent for MicroLED-integrated chip packaging, enabling optical interconnects via through-glass vias (TGV), though its practical impact remains speculative.
Security concerns persist: Anthropic’s report flags Zhipu AI’s GLM-5.3 as a frontier open-weight model capable of bypassing safeguards, rivaling Claude’s Mythos AI in generating harmful content. Nvidia launched its Open Agent Safety Platform, combining OpenShell sandboxing and BlueField-powered sentry hardware to monitor rogue agents. Meanwhile, Epoch AI claims AI costs are dropping ~50% per quarter, outpacing Moore’s Law in reducing intelligence expenses.
Model page: GLM-5.3 →
The story so far
3 episodes →- OpenAI’s Jalapeño ASIC beats Nvidia on efficiency in AI chip designthis story
This week on Tom's Hardware Premium: October 3, 2026
Tom's Hardware · 3 October 2026
Loading the full article…
This text was published by Tom's Hardware and written by Sayem Ahmed. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Hardware & Compute
All →- Meta releases open-source Muse Gadgets SDK and free Muse Home Link for U.S. subscribers · 6 src
- NVIDIA launches DGX Spark 64 GB desktop AI system starting at $4,999 · 11 src
- Nvidia raises Shield TV Pro price by $100 citing AI-driven component costs · 2 src
- gufo-Qwen3.6-35B-A3B-Q6dense - 3095tok/s prefill; 190 tok/s decode on Strix Halo · 2 src
- SpaceX launches Google AI chips into orbit for Project Suncatcher test · 12 src
Comments
via GitHub Discussions