TECHNOLOGY

OpenAI Unveils In-House 'Jalapeño' Chip Benchmarks Outperforming Nvidia Blackwell

PUBLISHED Aug 25, 2026, 12:54 PM ET

Read, Watch or Listen

Media Bias Meter

Sources: 6

Center 100%

Sources: 6

OpenAI on Tuesday published benchmark results for its first in-house artificial intelligence processor, codenamed Jalapeño, developed in partnership with Broadcom. The custom application-specific integrated circuit (ASIC) is engineered specifically for AI model inference rather than general-purpose training. Evaluated on SemiAnalysis’s InferenceX platform across public models including GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2.5 1T, Jalapeño achieved 1.5 to 1.9 times higher output per watt than prevailing commercial systems. On DeepSeek R1, the hardware reduced end-to-end latency by 1.7 to 3.6 times and produced over 700 tokens per second for low-concurrency user workloads. OpenAI confirmed that internal AI tools assisted in the chip’s physical layout, bring-up, and code optimization. Rated at 700 watts TDP with sustained drawing below 550 watts, deployment into primary infrastructure is scheduled for late 2026 alongside continued use of external accelerators.

By James Porter | JQJO News

Timeline of Events

On October 2025 OpenAI partnered with Broadcom to co-design custom computing accelerators.
On November 2025 engineers started nine-month physical tapeout process for custom silicon.
On June 24 2026 OpenAI publicly revealed development of custom Jalapeño inference silicon.
On August 25 2026 (19:00 UTC) OpenAI published official Jalapeño benchmark metrics on InferenceX.
On August 25 2026 (20:00 UTC) SemiAnalysis confirmed performance metrics and comparison data against Blackwell.
On August 25 2026 (21:00 UTC) industry analysts evaluated internal custom silicon economics versus merchant GPUs.
By December 2026 OpenAI expects first-generation Jalapeño chips to begin infrastructure integration.
By June 2027 initial volume deployment of Jalapeño servers will expand capacity.
By December 2027 second-generation OpenAI inference silicon is scheduled for deployment readiness.
By December 2028 multi-generation custom chip program targets ten gigawatts total compute.

News Intelligence

Immediate US impact: Enhances economic viability for running high-speed consumer AI applications.
Possible long-term US impact: Reduces cloud reliance on dominant single-vendor merchant GPU platforms.
Most affected groups: AI infrastructure engineers, chipmakers, enterprise software developers, enterprise buyers.
What readers should prioritise: Track deployment schedules, verified production yields, and total real-world efficiency.