OpenAI and Broadcom Forge $10B Partnership for Custom AI Chip
OpenAI, in a strategic move, is gearing up to launch its inaugural custom artificial intelligence chip in 2026. This venture is made possible through a partnership with the semiconductor behemoth, Broadcom, as confirmed by industry insiders. The deal, estimated at a whopping $10 billion, symbolizes OpenAI’s audacious bid to lessen its reliance on Nvidia, thereby lowering the costs of training and operating its GPT models.
Broadcom’s CEO, Hock Tan, disclosed during the company’s earnings call that an unidentified customer had placed orders exceeding $10 billion for custom AI accelerators. This customer was later revealed to be none other than OpenAI. The chips, internally dubbed as XPUs (eXtreme Processing Units), are intended to be used in-house initially, rather than being sold to external clients. This alliance has the potential to yield 1-2 million custom processors across thousands of racks, thereby challenging some of the largest AI inference clusters of today.
The announcement triggered a surge in Broadcom’s stock by 9.4% on Friday. With this, the company joins the league of Google, Meta, and ByteDance as major patrons for custom AI silicon. The shipments are slated to kick off in 2026, which could position OpenAI competitively against established entities. This could also potentially provide the company with an upper hand in future negotiations with AMD and Nvidia.
Source: CNBC