OpenAI’s Custom Jalapeño AI Chip Shows Major Gains in Speed and Energy Efficiency
OpenAI has published the first performance results for Jalapeño, its first custom inference chip, revealing significant gains in both speed and energy efficiency compared to existing hardware.
According to the company's internal benchmarks, Jalapeño delivered between 1.5 and 1.9 times more AI work per watt and substantially lower latency than comparison systems across multiple AI models, including GPT-OSS 120B, DeepSeek R1, and Kimi K2.5 1T.
The results mark a major milestone in OpenAI's long-term strategy to reduce its dependence on external chip suppliers and optimize its infrastructure for the unique demands of its AI models. The chip is part of a broader strategy covering chips, memory, networking, models, and AI products, signaling OpenAI's ambition to build a vertically integrated AI stack.
The Jalapeño chip is designed specifically for inference workloads, which are the process of running AI models to generate responses. While the chip is not intended for training, which requires even more computational power, its efficiency gains for inference could have a significant impact on the cost and scalability of AI services.
OpenAI has been exploring the development of custom chips for years to address the high cost of running its AI models on third-party hardware. The company is reportedly collaborating with major chip foundries and design firms to bring Jalapeño to production.
The new benchmark results suggest that the project is making substantial progress, potentially offering a roadmap for other AI labs to follow.
"Jalapeño represents a significant step forward in our efforts to build the most efficient AI infrastructure in the world. The combination of performance per watt and low latency is critical for delivering the best possible experience to our users." – OpenAI spokesperson
The successful development of Jalapeño could have far-reaching implications for the AI industry. By reducing its reliance on external chip suppliers, OpenAI could gain greater control over its infrastructure, potentially lowering costs and improving performance.
The move also puts pressure on other AI companies to consider developing their own custom chips to stay competitive. OpenAI's custom chip strategy is part of a broader trend of tech companies designing their own silicon to optimize for their specific workloads, potentially reshaping the supply chain for AI accelerators.
As OpenAI continues to develop and deploy Jalapeño, its impact on the broader AI ecosystem will be closely watched. The success of the chip could validate the strategy of custom silicon for AI and accelerate the trend of vertical integration, potentially reshaping the competitive dynamics of the AI industry.
This development follows a TechCrunch report that highlighted the growing interest among AI companies in custom silicon and a Reuters analysis of the strategic implications.
