OpenAI to Launch AI Chip with Broadcom, TSMC
OpenAI is reportedly collaborating with Broadcom on the development of a specialized AI chip designed to run trained artificial intelligence models, according to sources familiar with the matter. The AI-focused startup and the chip manufacturer are also in discussions with TSMC, the world's leading chip contract manufacturer. Sources indicate that OpenAI has been planning this custom chip for approximately a year, focusing on technology suited for these purposes, though discussions remain in preliminary stages.
Earlier reports from June suggested that Broadcom was exploring the possibility of manufacturing an AI chip for OpenAI. The design and production process for such chips is lengthy and costly, and OpenAI's current strategy is less focused on GPUs—NVIDIA's domain for training and building generative AI models. Instead, OpenAI is exploring a dedicated chip optimized for software processing and user response tasks, a process known as inference. Analysts and investors anticipate that as more tech companies use AI models for complex functions, demand for inference-supporting chips will rise.
A source notes that OpenAI may still consider building its own chip production network but recognizes that developing custom chips with partners is a faster and more viable approach. Recent reports suggest OpenAI has set aside plans to establish its own chip manufacturing facility.
Broadcom is among the largest designers of ASICs (application-specific integrated circuits), creating chips tailored for singular, client-specified applications. One of Broadcom's largest clients in this field is Google, part of Alphabet. The company also collaborates with Meta and ByteDance, the owner of TikTok.
When asked in September about potential new clients for AI training given the high demand, Broadcom CEO Hock Tan stated that they only add clients once projects reach a production-scale volume. He clarified on an earnings call, “This is not a product that any customer can deploy easily, and we don't consider proof-of-concept as mass production.”
OpenAI's services require substantial computing power for development and deployment, most of which currently relies on NVIDIA chips. To meet its needs, the industry has been seeking NVIDIA alternatives, including AMD’s AI accelerators and other custom options.
OpenAI is also actively planning investments in data centers where these AI chips will eventually be deployed. Company leadership has advocated for expanded data center capacity to the U.S. government. CEO Sam Altman has consulted with global investors, including some from the Middle East, to secure funding for these initiatives.
