OpenAI, a well-funded AI startup, is exploring the possibility of designing its own AI chips as the chip shortage for training AI models continues to worsen. Discussions regarding AI chip strategies have been ongoing within the company since at least last year. OpenAI is considering various strategies to advance its chip ambitions, including acquiring an AI chip manufacturer or launching an internal chip design initiative. OpenAI CEO Sam Altman has prioritized the acquisition of more AI chips for the company.
Presently, like many of its competitors, OpenAI relies on GPU-based hardware for developing models such as ChatGPT, GPT-4, and DALL-E 3. GPUs are suitable for training modern AI due to their ability to perform numerous computations in parallel. However, the surge in generative AI has strained the GPU supply chain significantly. Microsoft has faced such a severe shortage of server hardware for AI that it could result in service disruptions. Nvidia, a prominent GPU manufacturer, has reported that its high-performance AI chips are sold out until 2024.
Google, Amazon, and Microsoft have already introduced their proprietary AI chips
GPUs are also crucial for running and serving OpenAI’s models, which are operated in the cloud using clusters of GPUs. However, the cost of GPUs is substantial. An analysis by Bernstein analyst Stacy Rasgon estimated that if ChatGPT queries reached a scale similar to Google Search, it would require around $48.1 billion worth of GPUs initially and about $16 billion worth of chips annually to maintain operations.
OpenAI would not be the first organization to pursue the development of its own AI chips. Companies like Google, Amazon, and Microsoft have already introduced their proprietary AI chips for various AI-related tasks.
While OpenAI is well-positioned to invest heavily in research and development due to its substantial venture capital funding, the hardware business, especially AI chips, is highly competitive and capital-intensive. Past challenges and job cuts at AI chip manufacturers like Graphcore and Habana Labs highlight the complexities and uncertainties of the AI chip market. Developing a custom AI chip could take years and cost hundreds of millions of dollars annually, making it a risky endeavor even for well-funded companies like OpenAI.
