OpenAI Halts Foundry Plans, Will Instead Develop First In-House AI Chip with Broadcom, TSMC

California: In a bid to reduce its dependence on Nvidia and manage the rising cost of computes, OpenAI has allied with Broadcom and Taiwan Semiconductor Manufacturing Company to develop its first custom AI chip, putting on ice earlier plans to build out its foundry network.

The fast-emerging company best known for its AI-based ChatGPT, faces growing hardware costs as the company scales up operations. According to sources close to the company, OpenAI initially considered building a network of chip-manufacturing foundries but scaled it back due to prohibitive costs and timelines. Instead, OpenAI has decided to focus on in-house chip design, with Broadcom helping with the design process and TSMC providing manufacturing capabilities. The first bespoke chip is said to arrive in 2026, but the date may change.

To complement its infrastructure, OpenAI is also diversifying its chip supply, adding AMD’s new MI300X chips, supported by Microsoft’s Azure, to its current reliance on Nvidia. Although Nvidia still holds more than 80% of the AI chip market, shortages and increasing costs have led tech giants like OpenAI, Microsoft, and Meta to seek in-house or alternative solutions. AMD’s new AI chip lineup is expected to generate $4.5 billion in 2024 sales, challenging Nvidia’s dominance.

OpenAI has assembled a chip team of around 20 engineers, including former Google experts with experience building specialized AI chips, to lead its new project. Broadcom’s expertise in chip customization has been critical, particularly in designing parts that streamline data transfer in large-scale AI systems.

As OpenAI diversifies its chip sources, it remains careful not to poach talent from Nvidia. Instead, the company strives to maintain a good working relationship as it continues to depend on the latest chips by Nvidia for critical tasks. OpenAI’s custom chip initiative speaks to its drive to cut costs and establish stable access to AI infrastructure in an increasingly tight market where annual losses are expected to be $5 billion versus $3.7 billion in revenue.

Disclaimer: The news articles published on Innovate Marketers are based on reports from reputable third-party sources and are not original reporting by Innovate Marketers. While we strive to ensure accuracy and integrity, we cannot guarantee the completeness or timeliness of the information provided.