OpenAI and Broadcom have officially unveiled their debut custom-designed chip, codenamed “Jalapeño.” This marks a significant strategic entry for the ChatGPT developer into the highly competitive artificial intelligence silicon market. Manufactured by Broadcom, these chips are engineered to power OpenAI’s inference operations, the computationally intensive process of delivering its advanced AI models to users across ChatGPT and a growing ecosystem of applications.
In a candid conversation with CNBC’s David Faber, OpenAI President Greg Brockman revealed that the Jalapeño chip was developed in an accelerated timeline of just nine months, with a substantial assist from OpenAI’s own cutting-edge AI models. “The degree to which our models have been able to accelerate its development was truly remarkable, exceeding our initial expectations,” Brockman shared.
Broadcom, a key player in the semiconductor industry, has emerged as a significant beneficiary of the generative AI surge. The company has been instrumental in enabling hyperscalers and leading AI research labs to develop their own bespoke silicon solutions. This strategic partnership with OpenAI further solidifies Broadcom’s position in the AI hardware landscape, with its shares already showing strong upward momentum in 2026.
The announcement sent a positive ripple through the market, with Broadcom’s stock climbing on Wednesday. Brockman emphasized the sheer scale of OpenAI’s compute needs, stating, “We simply cannot acquire compute power fast enough.” This sentiment was echoed by Broadcom CEO Hock Tan, who described the demand from his enterprise clients, including OpenAI, as “simply insatiable.” Tan further elaborated, “It’s just much more than we can address, and this isn’t a fleeting trend; we are seeing this demand not only in ’26 and ’27 but also an elevated and sustained demand projected into ’28 as well.”
The Jalapeño chip represents a pivotal advancement in OpenAI’s ambitious strategy to “build the full stack behind its models and products.” Brockman articulated the vision behind this vertical integration: “By taking greater ownership of more components within our infrastructure, we can deliver enhanced intelligence with superior efficiency, thereby accelerating the path toward making advanced AI more broadly accessible.”
Since igniting the generative AI revolution in 2022, OpenAI has been a voracious consumer of high-performance computing, particularly NVIDIA’s potent Graphics Processing Units (GPUs). These GPUs have been the bedrock for training massive AI models and handling intensive computational workloads. However, the exponential growth in demand has necessitated diversification of OpenAI’s silicon sourcing strategy.
Earlier this year, OpenAI forged a significant cloud computing agreement with Amazon Web Services (AWS), which includes access to AWS’s proprietary Trainium AI chips. Additionally, OpenAI has secured partnerships with NVIDIA’s competitor, Advanced Micro Devices (AMD), and with Cerebras, a specialized AI chipmaker that recently completed its initial public offering. These multi-pronged approaches underscore OpenAI’s commitment to ensuring a robust and scalable compute infrastructure.
The collaboration between OpenAI and Broadcom was first disclosed in October, following an 18-month development period. The initial plan outlined the development and deployment of custom OpenAI-designed chips, with an ambitious target of generating enough computational power to require 10 gigawatts of energy.
The Jalapeño chip is classified as an Application-Specific Integrated Circuit (ASIC). While industry experts note that ASICs typically offer less flexibility compared to general-purpose GPUs like those from NVIDIA, they excel in cost-effectiveness and can be meticulously optimized for specific AI tasks. OpenAI’s rapid nine-month design cycle for Jalapeño, coupled with its involvement in designing significant portions of the supporting computer system, highlights the synergistic benefits of this partnership.
Positioned as an “Intelligence Processor,” the companies are marketing Jalapeño as the inaugural “AI accelerator” within a broader platform designed to “make advanced AI faster, more reliable, and more accessible to more people.”
A physical prototype of the new chip is slated for delivery to OpenAI imminently. Both companies are targeting an initial deployment of the Jalapeño chips by the end of 2026, with plans for significant expansion in the subsequent years. Tan elaborated on the deployment timeline, projecting “small prototype development” in late 2026, followed by a substantial ramp-up in 2027, leading to full-scale operations in the first half of 2028.
Original article, Author: Tobias. If you wish to reprint this article, please indicate the source:http://aicnbc.com/23126.html
