OpenAI Plans to Make Its Own Chips for Next-Gen AI Models

The makers of ChatGPT and GPT-4, OpenAI, are looking into the option of making their own AI chips for their future models. Reuters reports that the company is also considering buying some chip design companies, according to sources inside the firm. The final decision is not yet made, but OpenAI has been facing problems with the low supply and high cost of AI chips since last year, with NVIDIA as their main provider. OpenAI’s CEO, Sam Altman, has expressed his concern about the lack of GPUs, which are mostly controlled by NVIDIA, who has an impressive 80% share of the global market for AI-specific chips.

In 2020, OpenAI relied on a huge supercomputer built by Microsoft, who is a big investor in OpenAI, which uses 10,000 NVIDIA GPUs. This system is essential for running ChatGPT, which has a very high cost, according to Bernstein’s analyst Stacy Rasgon. Each ChatGPT interaction is estimated to cost about 4 cents. Comparing it with Google search, if ChatGPT ever reached a tenth of Google’s search volume, the initial GPU investment would jump to a staggering $48.1 billion, with an additional annual cost of around $16 billion for maintenance. OpenAI did not respond to any requests for comment. The possible move to custom silicon shows that OpenAI wants to be more independent and cost-efficient for its AI development.

Who is OpenAI?

OpenAI is an AI research and deployment company that aims to create safe and beneficial artificial general intelligence (AGI) for humanity. It consists of a non-profit organization and a for-profit subsidiary, and it is funded by several prominent investors and partners. It is known for its cutting-edge research and products, such as ChatGPT, GPT-4, DALL·E, OpenAI Codex, and OpenAI Five.