OpenAI and Broadcom Unveil Jalapeño: First Custom AI Inference Chip Cuts LLM Costs by 50%

OpenAI and Broadcom recently unveiled Jalapeño, their first custom-built Application-Specific Integrated Circuit (ASIC) chip specifically engineered for large language model (LLM) inference. This significant development, announced this week, is projected to reduce LLM serving costs by 50%, directly addressing a major financial challenge for OpenAI. The collaboration aims to enhance the efficiency and affordability of deploying advanced AI models. For broader context, explore our Top 100 AI Tools.

Jalapeño's Core Function and Cost Reduction

The newly introduced Jalapeño chip is a purpose-built ASIC designed to optimize the inference process for large language models. OpenAI President Greg Brockman confirmed on CNBC on June 24, 2026, that the chip delivers "real performance improvement on performance per watt and performance per dollar." This improvement is crucial for companies like OpenAI, which incur substantial compute-related expenses. The reported 50% reduction in serving costs could significantly impact the operational economics of running sophisticated AI systems.

Accelerated Development with AI

A notable aspect of the Jalapeño chip's development was its rapid progression from initial schematics to fabrication within just nine months. OpenAI attributes this accelerated timeline to its innovative approach of utilizing its own AI models in the chip design process. This internal application of AI for hardware development highlights a potential new paradigm for technological innovation, where AI tools contribute directly to the creation of their own underlying infrastructure.

Addressing OpenAI's Financial Landscape

The introduction of Jalapeño directly targets OpenAI's largest cost center. In 2025, OpenAI reported revenues of $13.07 billion but simultaneously posted a substantial $21 billion operating loss. A significant portion of this loss, $19.18 billion, was attributed to research and development costs, primarily driven by compute expenses. The company's infrastructure payments to Microsoft alone exceeded $10.59 billion. With an Initial Public Offering (IPO) planned for 2026, the Jalapeño chip represents a strategic move to improve the company's financial health by mitigating these high operational costs, making its services more sustainable and potentially more profitable.

Testing and Future Rollout

OpenAI has already commenced testing its GPT-5.3-Codex-Spark model on the new Jalapeño silicon. The company has outlined plans for data center rollouts of the custom chip to begin by the end of 2026. This phased deployment indicates a methodical approach to integrating the new hardware into its existing infrastructure, ensuring stability and performance as it scales its operations.

Strategic Chip Partnerships Remain

Despite the unveiling of its custom inference chip, OpenAI's existing agreements with other major chip and cloud providers remain intact. The company maintains significant deals, including a $30 billion agreement with Nvidia and a $50 billion agreement with AWS. Furthermore, partnerships with AMD and Cerebras also continue. This strategy suggests that while OpenAI is investing in custom silicon for specific optimizations, it is also maintaining a diversified approach to its compute infrastructure, leveraging a range of technologies and providers to meet its diverse needs.

Conclusion

The joint unveiling of the Jalapeño custom AI inference chip by OpenAI and Broadcom marks a pivotal moment in the pursuit of more cost-effective and efficient large language model operations. By reportedly cutting LLM serving costs by 50% and leveraging AI in its own design, Jalapeño addresses OpenAI's substantial compute expenses ahead of its planned 2026 IPO. The ongoing testing with GPT-5.3-Codex-Spark and planned data center rollouts by late 2026 underscore a strategic effort to enhance performance and financial sustainability, while existing partnerships ensure a robust and diversified infrastructure.

Sources

• https://openai.com/index/openai-broadcom-jalapeno-inference-chip/
• https://venturebeat.com/infrastructure/openai-unveils-first-custom-ai-inference-chip-jalapeno-with-broadcom-and-its-development-was-sped-up-with-openais-own-models
• https://investors.broadcom.com/news-releases/news-release-details/openai-and-broadcom-unveil-llm-optimized-intelligence-processor
• https://www.cnbc.com/video/2026/06/24/openai-president-greg-brockman-on-new-chip-this-is-a-real-performance-improvement.html