OpenAI Jalapeño Chip Offers Efficiency Gains Over Nvidia
The custom-designed inference silicon hits a 50 percent efficiency target ahead of a larger volume rollout in 2027.
Updated on Oct. 16, 2026 in Semiconductors
Live Poll
Should large tech companies prioritize designing their own hardware to improve performance and cost efficiency?
TIME has recognized the OpenAI Jalapeño chip, an application-specific integrated circuit designed for inference, as one of its 2026 Best Inventions. The custom silicon, announced in June 2026, aims to optimize AI cost and power efficiency.
Why it matters
The development represents a strategic shift toward custom silicon to lower cost-per-token and energy consumption metrics by a targeted 50 percent. This hardware push allows for greater control over performance as infrastructure demands scale toward a 10-gigawatt target by 2029.
The Jalapeño delivers 13.4 petaflops of 4-bit compute capacity alongside 232 GB of memory and 15.4 TB/s of bandwidth. It sustains power at or below 550 W, with a maximum rating of 700 W.
The players
OpenAI
An AI research and deployment organization focused on scaling large language models and vertical integration of hardware and software.
Broadcom
A global semiconductor firm known for designing custom ASICs and high-speed networking components for large-scale data centers.
Celestica
A supply chain and manufacturing services company providing complex hardware engineering and production support.
The details
The chip is a specialized application-specific integrated circuit (ASIC—a processor built for a single, specific task rather than general computing) optimized exclusively for inference. Developed in collaboration with Broadcom and Celestica, the design moved from initial architecture to tape-out—the final stage of design before manufacturing—in nine months. The silicon leverages high-bandwidth memory architecture to feed its 4-bit compute units, directly targeting the energy-intensive processing required for large-scale AI workloads.
Timeline
October 2025: OpenAI and Broadcom partnership announced.
June 24, 2026: OpenAI unveiled the Jalapeño chip.
August 25, 2026: Benchmark results published.
October 15, 2026: Jalapeño named to TIME Best Inventions list.
Late 2026: Small-volume deployment planned.
The Tech Race
The Jalapeño chip positions OpenAI alongside hyperscale cloud providers currently migrating to proprietary inference silicon. This move marks a departure from reliance on general-purpose hardware toward vertically integrated stacks designed to maximize tokens-per-watt.
Small-volume deployment of the hardware is scheduled for late 2026, with a wider ramp-up expected throughout 2027. While not available for individual purchase, the chip's efficiency gains are intended to lower operational costs for AI services.
The takeaway
The Jalapeño demonstrates that custom silicon development cycles can reach tape-out in under a year, signaling a rapid acceleration in vertical integration. Observers should monitor the transition from small-volume deployment in late 2026 to the broader capacity rollout planned for 2027.
Further reading
For broader trends in custom processor development, see our Semiconductors section.
Source note: This article includes information reported by Crypto Briefing.
Live Poll
Should large tech companies prioritize designing their own hardware to improve performance and cost efficiency?
