OpenAI is collaborating with Broadcom and Taiwan Semiconductor Manufacturing Company (TSMC) to develop its first in-house chip aimed at enhancing its artificial intelligence infrastructure.
The initiative comes as OpenAI diversifies its chip sourcing by incorporating AMD's MI300X chips alongside NVIDIA's market-leading GPUs. Previously, OpenAI considered a $7 trillion investment to establish its own chip foundries but pivoted to design chips with established partners due to the project's complexity and cost.
With increasing demand for AI computing power, OpenAI's strategy aligns with industry leaders like Google and Amazon, focusing on a blend of internal chip design and external sourcing. The custom chip, expected to launch in 2026, is designed primarily for inference tasks, which will facilitate the real-time application of AI models.
Currently, OpenAI relies heavily on NVIDIA, which commands over 80% of the AI chip market. However, supply chain issues and rising costs have prompted OpenAI to seek alternative partnerships. AMD, anticipating $4.5 billion in AI chip sales for 2024, is supplying OpenAI through Microsoft's Azure platform.
Despite projecting revenues of $3.7 billion for 2024, OpenAI expects a $5 billion loss due to significant hardware, electricity, and cloud service costs. By broadening its supply chain, the company aims to manage expenses and optimize resources as it scales its AI offerings, including ChatGPT.