OpenAI Unveils Jalapeño AI Chip to Challenge Nvidia Dominance

OpenAI has unveiled its first custom inference processor, Jalapeño, which the company claims outperformed Nvidia’s GB300 accelerators in power efficiency and response speed during internal testing [1, 2]. While the move signals a strategic shift toward vertical integration, OpenAI confirmed it will continue to rely on Nvidia hardware for both training and inference workloads [1, 2].

Performance and Power Efficiency

The Jalapeño chip is designed specifically for inference—the process of running trained AI models to generate responses—rather than the more intensive task of model training [2, 3]. In tests conducted by OpenAI using the public InferenceX benchmark, the 700-watt processor demonstrated 1.5 to 1.9 times higher AI work per watt at peak throughput compared to Nvidia’s GB200 and GB300 systems [1]. For interactive workloads, OpenAI reported performance gains of 2.1 to 4.1 times over the Nvidia hardware [1].

OpenAI chip chief Richard Ho noted that the chip’s measured power consumption remained at or below 550 watts during testing, even though the hardware is rated at 700 watts [1, 2]. Because the benchmark calculations used the full 700-watt rating, the actual efficiency gains may be higher than the published figures [1]. The tests utilized open-source models, including DeepSeek R1 670B and Kimi K2.5 1T, to validate the chip's capabilities [1, 2].

Market Strategy and Future Roadmap

Despite the benchmark results, OpenAI’s announcement included a commitment to maintain its existing supply chain with Nvidia [1]. Nvidia remains a dominant force in the sector, reporting $75.25 billion in data center revenue for the first quarter of fiscal 2027, a 92% increase year-over-year [1]. OpenAI’s decision to develop Jalapeño with Broadcom is primarily a cost-containment program intended to lower infrastructure expenses as the company scales its operations [2, 4].

OpenAI plans to begin deploying Jalapeño within its own data centers by the end of the year [1, 2]. The company has already initiated development on second- and third-generation processors, aiming to further optimize the cost and power requirements of its global AI infrastructure [2].

What to watch

• Deployment Timeline: Monitor the rollout of Jalapeño chips into OpenAI’s production infrastructure, expected by the end of the year [1, 2].
• Next-Gen Development: Watch for updates on the second-generation chip, which is expected to reach the tape-out stage in the coming months [2].
• Nvidia Purchasing Patterns: Observe whether OpenAI’s reliance on Nvidia shifts as Jalapeño enters mass deployment, or if the company maintains its current purchasing volume to support training needs [1, 2].

The long-term impact of Jalapeño depends on OpenAI’s ability to scale production and realize the projected infrastructure savings. For now, the chip serves as a specialized tool for inference efficiency rather than a total replacement for Nvidia’s broader ecosystem [1, 2].