AI company OpenAI plans to collaborate with U.S. semiconductor giant Broadcom to mass-produce its self-developed AI chips. Sources indicate that the co-designed chip is expected to be delivered next year. The partnership signals the tech industry’s growing interest in seeking alternatives to customized NVIDIA chips.
Broadcom CEO Hock Tan revealed last Thursday that an undisclosed new client had placed orders totaling up to $10 billion with the company. Although Broadcom did not explicitly name the client, people familiar with the matter confirmed that it is OpenAI. Following the new announcement, Broadcom’s stock surged nearly 11% last Friday.
OpenAI’s move aims to address the enormous demand for computing power and reduce its reliance on NVIDIA. For months, OpenAI CEO Sam Altman has emphasized that GPU shortages have delayed the release of new versions of ChatGPT. The company heavily depends on NVIDIA’s off-the-shelf GPUs, but soaring demand for training complex AI models has strained supply chains.
Amid these challenges, customized chips are gaining increased attention, with Broadcom holding a strong position in this field. OpenAI and Broadcom have been working together for over a year to develop chips tailored for model training. Broadcom specializes in what it calls “XPU” semiconductors, which are designed for specific applications such as training AI models like ChatGPT. Analysts at HSBC recently projected that Broadcom’s customized chip business will grow significantly faster than NVIDIA’s by 2026. However, The Wall Street Journal reported that Broadcom’s chips are not meant to challenge NVIDIA but rather to help meet OpenAI’s hardware needs.
OpenAI’s strategy aligns with that of other tech giants like Google, Amazon, and Meta, all of which have developed their own proprietary chips for AI applications. Demand for computing power to train and run AI models continues to surge. An analyst from eMarketer noted that while NVIDIA’s GPUs remain the go-to choice for AI companies, custom chips can enhance specific performance and help overcome bottlenecks.
Separately, The Information reported that OpenAI has significantly raised its projected cash burn for the coming years. The latest forecast suggests the company will consume $115 billion in cash by 2029, $80 billion higher than previous estimates. This year, its cash burn is expected to exceed $8 billion, $1.5 billion more than initially projected. To manage rising costs, OpenAI will also pursue developing its own data center server chips.
