OpenAI close to developing its own chip for optimized AI tasks

OpenAI is rapidly developing its chip for Artificial Intelligence tasks. The ChatGPT creator is working with Broadcom while sourcing chips from NVIDIA and AMD.

OpenAI to rely on Broadcom to develop its own AI chip

Generative Artificial Intelligence apps, services, and platforms are growing at an unprecedented pace. One of the fastest-growing companies in the AI space is OpenAI.

Similar to other AI startups, even the Microsoft-backed company is spending millions to optimize AI. While chipmakers like AMD and NVIDIA are churning out processors optimized for AI workloads, OpenAI is confident it can build a custom AI chip.

According to Reuters, OpenAI has revised its hardware strategy. It is now working with Broadcom to design an AI chip best suited for OpenAI’s workloads. In other words, Broadcom is OpenAI’s custom silicon partner.

Multiple reports have suggested that OpenAI was actively scouting for chip manufacturers. While companies such as AMD, NVIDIA, Samsung, TSMC, and others are quite busy, Broadcom may have reserved manufacturing for OpenAI.

When will Broadcom start manufacturing AI chips for OpenAI?

Before looking for partners, OpenAI reportedly considered designing and manufacturing chips. The company may have even attempted to raise capital to build foundries.

OpenAI may have figured out that building factories and manufacturing chips is quite complex, time-consuming, and expensive. No wonder, established companies are trying to Intel’s foundries as Intel has laid the groundwork and developed the manufacturing technology over many years.

To meet its immediate demand, OpenAI is sticking with established companies such as AMD and NVIDIA. However, OpenAI is leveraging industry partnerships to optimize costs and lower manufacturing times. The company is also adopting a mix of internal and external agencies to secure chip supply.

Sources who divulged this information, have further indicated that OpenAI has been working with Broadcom for months. Interestingly, apart from “training chips”, OpenAI could also rely on Broadcom for “inference chips”.

Some reports suggest the first AI chip designed by OpenAI and manufactured by Broadcom could arrive in 2026. However, given the complexities involved in designing, testing, optimizing, and mass manufacturing chips, the timeline could shift.