OpenAI, which develops ChatGPT, is reported to be moving to develop its own chip

Reuters reports that OpenAI, which develops chatbot AI ChatGPT and image generation AI Dall-E, is considering developing its own artificial intelligence chip, and is also considering acquiring semiconductor companies.

Exclusive: ChatGPT-owner OpenAI is exploring making its own AI chips | Reuters
https://www.reuters.com/technology/chatgpt-owner-openai-is-exploring-making-its-own-ai-chips-sources-2023-10-06/

OpenAI has developed a large-scale language model called ``GPT'', and ChatGPT, which is based on this GPT, has become a hot topic. In July 2023, the API of GPT-4, the latest model at the time of article creation, will be released to the public.

OpenAI makes ``GPT-4'' publicly available for anyone to use, and officially releases the function to execute code within ChatGPT - GIGAZINE

In addition, Microsoft is the biggest supporter of OpenAI, and Microsoft's Bing Chat is a GPT-based chatbot AI, as well as the image generation AI DALL-E 3 .

Microsoft's official feature that allows you to easily generate anime-style illustrations and live-action images for free has been added to Bing chat, so I tried using it, and OpenAI's DALL-E 3 can generate ultra-high-definition images - GIGAZINE

Developing and operating such AI requires a huge amount of time and effort, and requires investment not only in software but also in hardware. OpenAI is developing AI using a supercomputer built by Microsoft using NVIDIA GPUs.

Microsoft spends tens of billions of yen to build a supercomputer for AI development for OpenAI, the developer of ChatGPT - GIGAZINE

The cost per query for ChatGPT is approximately 4 cents (about 6 yen). If ChatGPT were to grow to about one-tenth the size of Google Search, the cost would rise to $48.1 billion (approximately 7.15 trillion yen), and it would take $16 billion (approximately 2.2 trillion yen) per year to maintain operation. 370 billion yen) will be required.

However, the selling price of NVIDIA's Tensor core GPU 'NVIDIA H100' used for AI development has soared due to the intensification of AI development competition in recent years, and the cost is expected to increase further in the future.

Why is there a shortage of high-end GPU 'NVIDIA H100', which is coveted for the development of generative AI and large-scale language models? -GIGAZINE

Therefore, according to Reuters, OpenAI is considering developing its own chip for AI development, and is also considering acquiring semiconductor companies. However, OpenAI is still in the consideration stage and has not actually decided to develop the chip, and it is possible that it will take a policy of collaborating more closely with semiconductor companies including NVIDIA. Furthermore, even if OpenAI were to give the go-ahead to plans to develop its own chips, including acquisitions, it is likely that it would take several years for the plan to actually take shape, and during that time it would have to rely on chips from NVIDIA and AMD. It will be.

In addition, Meta, which does not have a semiconductor business and focuses on AI development, has been developing its own chips for many years, but it is reported that it is having a difficult time.