© ROOT-NATION.com - Use of content is permitted with a backlink.
OpenAI has unveiled its first specialized inference processor, developed and built in collaboration with Broadcom. The new chip is called Jalapeño and was designed specifically to meet the needs of OpenAI’s inference systems. The company also reported that its own artificial intelligence models were used in the processor’s development, and that it took less than a year to bring Jalapeño from initial design to production.
Read also: AERONAUT – everything that flies above the ground: aviation, UAVs and drones, rockets, and space
Although the chip is still undergoing testing, according to OpenAI, preliminary results show significantly better performance per watt compared to the most advanced alternatives currently available.
The partnership with Broadcom was officially announced back in October, but OpenAI had long had plans to develop its own chips to reduce its reliance on NVIDIA graphics processors. Google and Amazon are already following this same path – they have also developed their own specialized processors to accelerate machine learning tasks. OpenAI designed the chip itself, while Broadcom was responsible for silicon implementation and networking technologies.
Read also: A new level of teamwork: Anthropic introduces the AI assistant Claude Tag for Slack
OpenAI President Greg Brockman explained the company’s approach to processor development in an episode of the corporate podcast following the announcement of the partnership with Broadcom. “We have a deep understanding of what these workloads are,” Brockman said. “We were really looking for specific workloads that aren’t getting enough attention, [and we asked ourselves]: How can we build something that pushes the boundaries of what’s possible?”
Read also: How Wearables and AI Glasses Are Rewriting the Logic of Technological Civilization
Jalapeño is specifically designed for inference – the process of running pre-trained AI models in response to user queries. In its launch announcement, OpenAI highlighted the chip’s low operating cost when working with models for real-time programming.
Tasks that require more resources, such as pre-training models, will likely continue to be performed on NVIDIA hardware. However, even a modest reduction in inference costs could have a positive impact on financial performance. The company plans to deploy the Jalapeño platform in production systems starting in late 2026.
Optimizing inference systems could become one of the key factors in the future economics of artificial intelligence. This process is expected to span all levels of the technology stack. OpenAI is already developing agent-based products, such as Codex, as well as the models on which they run and the data centers required to deploy them. Developing its own specialized chips allows the company to integrate all infrastructure components even more deeply.
Read also:
- XR headsets and glasses set to reach new heights: Qualcomm unveils Snapdragon Reality Elite
- OpenAI expands ChatGPT’s capabilities: What the new Scheduled Tasks system can do