Key to Cost and Supply Chain Manangement

OpenAI considered building its own fab to reduce long-term costs and control the supply chain, but ultimately chose to outsource production due to the huge investment and time risk. The company now works with Broadcom (AVGO) and TSMC (TSMC) to focus on the design and development of inference chips to reduce its reliance on expensive GPUs.

OpenAI's strategy reflects the model of large technology companies such as Amazon (AMZN), Meta (META), Google (GOOGL) and Microsoft (MSFT), which aims to reduce cost pressure and improve supply chain flexibility through a mix of internal and external chip supply.

Through a diversified supply chain strategy, OpenAI reduces the constraints of a single supplier on its infrastructure, thereby enhancing cost management and supply stability.

Competition in Chip Market Intensifies

The current market demand for training AI models is still dominant, but as AI applications continue to expand, the demand for inference chips is growing rapidly. To this end, OpenAI has formed a team of 20 senior engineers dedicated to the development of inference chips, including engineers who have worked at Google and brought professional experience in Google's self-made TPU.

Currently, OpenAI has booked TSMC's production capacity for 2026 to ensure supply capabilities. Although production plans may be adjusted, this order still shows the initial success of OpenAI's AI chip development strategy.

Currently, Nvidia (NVDA) GPU has a market share of about 80%, but as costs rise and supply shortages occur, large companies such as OpenAI, Microsoft and Meta are beginning to seek alternatives, and OpenAI's reasoning chip development may pose a challenge to the market dominated by Nvidia.

Combining AMD Chips and Microsoft Azure

In addition to developing its own chips, OpenAI also plans to use AMD's new MI300X chip through Microsoft's Azure cloud service. The chip focuses on AI reasoning and training operations and aims to seize Nvidia's market share. AMD expects the MI300X chip to generate $4.5 billion in revenue in 2024, indicating that AMD's cooperation with Microsoft may become the main hardware choice for companies such as OpenAI in the near future.

By using AMD chips on Azure, OpenAI can reduce its reliance on expensive GPUs, improve computing cost-effectiveness, enhance computing infrastructure, and reduce the risk of dependence on a single supplier.

Continuously High Computing Costs Longing for Optimization

Informed sources revealed that OpenAI spends huge amounts of money every year on training AI models and maintaining the operation of applications such as ChatGPT. It is estimated that in 2023, OpenAI's revenue will be US$3.7 billion, but it may face losses of up to US$5 billion, with major expenditures concentrated on hardware, electricity and cloud services.

Therefore, OpenAI is continuously optimizing its computing architecture and supply chain to reduce costs, and plans to expand its suppliers from Nvidia to more partners. At the same time, in order to ensure a stable supply of Nvidia's latest generation of Blackwell chips, OpenAI is cautious about poaching Nvidia employees, aiming to balance long-term partnerships with the need for independent development, highlighting its delicate balance between pursuing chip autonomy and cooperative stability.