Exclusive Report: OpenAI to Launch First In-House AI Chip with Broadcom and TSMC

ChatGPT maker OpenAI to build its first in-house AI chip with Broadcom and TSMC

OpenAI is collaborating with Broadcom and Taiwan Semiconductor Manufacturing Company (TSMC) in developing its initial proprietary chip tailored to support its artificial intelligence (AI) systems. This initiative also involves integrating AMD chips alongside Nvidia chips to address the escalating infrastructure requirements, as disclosed by sources to Reuters.

Diversification of Chip Supply

As the prominent company behind ChatGPT, OpenAI has explored various avenues to diversify chip supply chains and minimize expenses. The consideration of constructing all components in-house and acquiring capital for an elaborate network of chip manufacturing facilities has been contemplated by OpenAI. However, due to the substantial costs and time required for such an endeavor, the company has currently set aside the ambitious foundry plans and shifted its focus towards internal chip design efforts, according to undisclosed sources.

Strategic Alliances and Cost Management

OpenAI's strategy underscores the Silicon Valley startup's approach of leveraging industry collaborations and a blend of internal and external methodologies to ensure chip availability and control costs, aligning itself with major industry players like Amazon, Meta, Google, and Microsoft. Being a significant chip purchaser, OpenAI's decision to engage with a diversified range of chip manufacturers while progressing with its customized chip development could have broader implications within the tech sector.

Broadcom observed a stock surge on Oct 29 subsequent to the news, with a closing increase of over 4.5 per cent, while AMD shares concluded the day with a 3.7 per cent uptick.

Focus on AI Capabilities

OpenAI, renowned for its role in commercializing generative AI for producing human-like responses, heavily relies on substantial computing power to train and operate its systems. With Nvidia's GPUs being a core component of its infrastructure, OpenAI employs AI chips for both training models and inference tasks. The company has been collaborating with Broadcom to develop its inaugural AI chip emphasizing inference capabilities.

Future Prospects

OpenAI is currently evaluating the potential development or acquisition of additional elements for its chip design and may extend partnerships, as detailed by sources. The company has curated a chip team comprising about 20 individuals, led by distinguished engineers with prior experience in building tensor processing units at Google, which expedite machine learning tasks.

Through Broadcom, OpenAI has secured manufacturing capacity at TSMC for fabricating its inaugural custom-designed chip anticipated to debut in 2026. However, the timeline for this release is subject to change.

Amid Nvidia's dominant market share exceeding 80 per cent, supply shortages and escalating costs have prompted major customers like Microsoft, Meta, and now OpenAI to explore internal or external alternatives.