OpenAI is preparing to enter the semiconductor race with its first in-house artificial intelligence chip, co-designed with US chipmaker Broadcom.
The move comes amid soaring demand for computing power to train and run large AI models, including OpenAI’s own GPT-5. According to people familiar with the matter, shipments are scheduled for next year.
Broadcom chief executive Hock Tan highlighted the arrival of a major new customer with $10bn in orders, without naming the firm.
However, multiple sources confirmed that OpenAI is the client, signalling a major shift in the AI hardware market dominated by Nvidia.
OpenAI and Broadcom partnership reshapes AI chip market
Broadcom has been building its custom AI chip business with large technology partners, previously working with Google on its Tensor Processing Units (TPUs).
The OpenAI deal, identified as Broadcom’s fourth major custom chip client, adds immediate scale. Tan told analysts that the agreement had brought “fairly substantial demand” and would begin shipping chips for the customer strongly from next year.
While Broadcom does not disclose customer identities, reports confirmed OpenAI’s role. Last year, both companies began collaborating on initial designs, though the production timeline had been unclear until now.
The chips, often referred to as “XPUs” to distinguish them from Nvidia and AMD’s GPUs, are positioned as alternatives to general-purpose processors that currently dominate AI workloads.
Rising competition for Nvidia in AI infrastructure
Nvidia remains the market leader in AI hardware, with hyperscale companies still representing a large share of its customer base. However, growth has slowed compared with the early surge of AI investment.
OpenAI was one of Nvidia’s earliest and biggest customers, relying heavily on its GPUs to scale ChatGPT and other models.
The arrival of Broadcom-designed XPUs for OpenAI signals that demand is beginning to diversify. HSBC analysts wrote ahead of Broadcom’s earnings that the company’s custom chip division is expected to outpace Nvidia’s GPU growth by 2026.
Shares of Broadcom have already risen more than 30% this year, with a further 4.5% jump in after-market trading following news of the new customer deal and stronger-than-expected earnings.
AI chip demand tied to OpenAI’s GPT-5 expansion
OpenAI chief executive Sam Altman has previously emphasised the urgent need for more computing capacity. Last month, he said the company would prioritise compute in light of “increased demand from GPT-5”, adding that OpenAI planned to double its compute fleet within five months.
The new chips will reportedly be used internally rather than sold to external clients, focusing on powering OpenAI’s own models and applications.
Industry-wide, the demand for AI compute continues to reshape infrastructure investment. Tech giants such as Google, Amazon, and Meta have already developed their own in-house chips.
OpenAI’s partnership with Broadcom now places it among those pursuing custom hardware strategies to control costs and meet rising performance needs.
The post OpenAI to launch first custom AI chip in 2025 with $10bn Broadcom partnership appeared first on Invezz