OpenAI custom chip project aims to challenge Nvidia's dominance

OpenAI designs a custom AI chip with Broadcom to rival Nvidia's market lead.

: OpenAI is developing a custom AI chip to challenge Nvidia's dominance in AI hardware. Led by Richard Ho, the 40-person team collaborated with Broadcom for the design. The chip design will be sent to TSMC for validation in upcoming months, with mass production expected in 2026. Initially for AI inferencing, it aims for a disruptive impact in the market.

OpenAI is taking strides to challenge Nvidia's stronghold on the AI hardware market by nearing the launch of its custom-designed AI chip. This effort is led by a small in-house team under Richard Ho, working in collaboration with Broadcom, known for its custom ASIC solutions. The design phase has reached the tape-out stage, with plans to send it to TSMC for validation soon, setting the stage for mass production in 2026.

The chip is designed to handle both AI model training and running, but initial applications will be focused on AI inferencing tasks. TSMC will utilize its 3nm technology node for manufacturing, integrating high-bandwidth memory typical of AI silicon designs. Though currently playing a minor role in OpenAI's infrastructure, the chip could significantly disrupt the market once fully deployed.

Despite the small current workforce, OpenAI sees this chip project as a pivotal tool for strategic moves in AI's future landscape. The tech company is already planning more advanced versions, preparing for broader adoption. As Nvidia holds around 80% of the market, other tech giants are also working on custom solutions, highlighting the intensely competitive race in AI acceleration.

OpenAIAI hardwareNvidia