OpenAI's Secret Weapon Jalapeño Beats Nvidia in Speed Tests, Will This Finally Lower Your ChatGPT Bill?
The new chip aims to slash costs and power usage for AI models.
The race to master artificial intelligence has hit a new hardware milestone. OpenAI has announced the development of its first custom chip named Jalapeño. This move comes as the company grapples with the enormous cost of running large language models. For tech consumers, this is a significant development because the price of computing power directly impacts the subscription fees for AI tools.
According to a report from Bloomberg, OpenAI hardware chief Richard Ho stated that the new processor outperforms Nvidia's GB300 in key performance metrics. The comparison focused on the amount of AI work completed per unit of power and the speed at which systems return responses. The results suggest that OpenAI is actively working to reduce the energy intensity of its infrastructure. This is not just a technical exercise but a financial necessity for a company facing rising utility costs.
The Jalapeño chip was developed with Broadcom. The two companies announced a multigeneration custom accelerator program last year. OpenAI revealed that the first design went from the initial concept to manufacturing tape-out in just nine months. This rapid development cycle is a testament to the speed of modern semiconductor engineering. It also highlights the urgency of building in-house capabilities to reduce reliance on external suppliers.
OpenAI is designing Jalapeño specifically for inference. This is the stage that happens after an AI model has been trained. During inference, the model responds to users generates content or performs tasks. The chip is optimized to handle high throughput, allowing it to serve large numbers of users efficiently. It also prioritizes low latency, which translates into faster response times for the end user. In the lab, the results are promising for both speed and efficiency.
The specific benchmark numbers released by OpenAI are quite striking. When tested against systems like GPT-OSS 120B and DeepSeek R1, Jalapeño delivered 1.5 to 1.9 times more AI work per watt of power. The chip also showed between 1.7 and 3.6 times lower end-to-end latency compared to the comparison systems. For highly interactive workloads, the company reported performance improvements of 2.1 to 4.1 times. These metrics indicate a potential shift in the efficiency landscape for data centers.
Power consumption is another major selling point for the new processor. While Jalapeño is rated at 700 watts, measurements presented at the Hot Chips conference showed sustained power at or below 550 watts for the workloads tested. OpenAI believes these efficiencies could significantly lower the cost of operating its rapidly expanding data center infrastructure. This is crucial because energy costs are one of the highest operating expenses in the tech industry today.
However, there are limits to the comparison. Jalapeño was not tested against Nvidia's newer Vera Rubin generation, which has already begun shipping. Additionally, the chip is not intended to train AI models, an area where Nvidia remains the dominant player. OpenAI is not preparing to abandon its partnership with Nvidia. Ho told Bloomberg that Nvidia remains a key partner and that OpenAI expects to continue buying large quantities of its processors as demand for computing power grows.
The bigger picture is that OpenAI is increasingly designing its own technology alongside the chips it buys from outside suppliers. The company says Jalapeño was created using knowledge from the models, software, networking and serving systems it operates. OpenAI's own AI models are also helping accelerate the chip's development. A second-generation processor is already well into development, with tape-out expected in the coming months. Conceptual work on a third generation is also underway. This vertical integration is the first step in bringing infrastructure costs down to a level that may eventually benefit consumers.
First reported by ibtimes.com.
More in Business
- IMF Chief Warns of Economic Divergence and Debt Risks Amid Artificial Intelligence Boom
Kristalina Georgieva suggests that while AI could significantly boost global growth, it simultaneously threatens to widen inequality and exacerbate inflationary pressures and public debt burdens.
- California Limits Utility Profits and Expands Home Energy Role
Gov. Gavin Newsom signed laws capping utility returns on lower-risk projects and accelerating virtual power plant development.
- Lyft to Pay $272.5 Million in Record California Wage-Theft Settlement
Largest such settlement in state history resolves claims that Lyft misclassified drivers as independent contractors from 2016 to 2020, denying them core wage protections.