OpenAI’s custom chip cuts AI costs in half with first "Intelligence Processor"
OpenAI’s new Jalapeño chip is more than a processor—it’s a financial lifeline. Designed in partnership with Broadcom, this custom application-specific integrated circuit (ASIC) targets the rising cost of running large language models (LLMs), a burden that last year alone consumed $8.4 billion of OpenAI’s budget. With weekly users now numbering 900 million, those operational expenses are projected to hit $14 billion in 2025. By building hardware tailored to its own models, OpenAI aims to shrink its infrastructure costs and regain control over a market currently dominated by third-party vendors.
A chip built for inference, not general AI
Unlike traditional AI accelerators, the Jalapeño chip is engineered specifically for LLM inference—the real-time processing that powers user interactions with models like ChatGPT. OpenAI provided the core architecture, while Broadcom handled the silicon design and high-speed networking integration. Early samples are already running unreleased workloads, including an internal GPT-5.3-Codex-Spark model, at target power and efficiency levels. Richard Ho, head of OpenAI’s hardware program, emphasizes that the design minimizes unnecessary data movement, pushing performance closer to theoretical limits.
Vertical integration: from silicon to software
This isn’t just about faster chips—it’s about full-stack control. OpenAI’s strategy mirrors Apple’s approach to hardware-software synergy, covering everything from chip design to memory systems, network scheduling, and final application layers. By aligning infrastructure with its internal model roadmaps, OpenAI can optimize efficiency at every stage. The result? Lower serving costs, more responsive products, and a self-reinforcing cycle of reinvestment into next-generation hardware.
Breaking the late-mover barrier
Competitors like Google have spent nearly a decade refining proprietary chips, giving them an early lead in AI hardware. But OpenAI’s move into custom silicon signals a shift: efficiency now trumps legacy advantages. With the Jalapeño chip, the company is betting on vertical integration to outpace rivals still reliant on third-party solutions. The financial stakes are high—OpenAI has committed $1.4 trillion to computing power over eight years—but the potential to redefine AI economics could be even greater.
Source: AI News. AI-assisted editorial synthesis — TechnoExpress.