Due to Massive AI GPU Construction Costs
"At Least 130 Trillion Won Needed to Replace Google Search"
American company OpenAI, which led the generative AI boom through the AI chatbot 'ChatGPT,' is reportedly facing massive losses. This is due to the computing costs and research and development (R&D) expenses required to develop large language model AI.
According to the U.S. media outlet The Information, OpenAI's operating loss last year was $540 million (approximately 720 billion KRW), about twice the amount from the previous year.
Since OpenAI is a private company, it is not obligated to disclose corporate information to the Securities and Exchange Commission (SEC) every quarter like other publicly traded U.S. companies. However, the media cited sources familiar with the company's internal affairs to obtain this information.
OpenAI launched a paid subscription version of ChatGPT in February and has been selling enterprise APIs since March. It also has major corporate clients such as investment bank Morgan Stanley and multinational social networking service Snap.
Internally, OpenAI expects revenue of $200 million (approximately 260 billion KRW) this year and $1 billion (approximately 1.3 trillion KRW) next year, but to turn the current massive losses into profits, it seems they will need to generate much more income.
ChatGPT Needs at Least '130 Trillion KRW' to Replace Google
The reason OpenAI, which sparked the 'ChatGPT boom,' is still struggling with losses is paradoxically due to the popularity of its AI product. Because the service cost of ChatGPT is so high, the more the service expands, the more the company loses money.
Regarding this, the U.S. media outlet The Washington Post (WP) reported in detail on March 5 (local time) that "the company loses money every time the AI chatbot speaks," covering ChatGPT's operating costs.
Artificial intelligence uses massive supercomputers made up of Graphics Processing Units (GPUs). Photo by Meta's supercomputer. [Image source=Yonhap News]
The media quoted Dylan Patel, senior analyst at U.S. semiconductor analysis firm Semianalysis, saying, "One line generated by ChatGPT can cost up to 1,000 times more than a simple Google search."
In particular, ChatGPT consumes a large number of Nvidia graphics processing units (GPUs), which are computer chips necessary for AI inference tasks. According to Patel, OpenAI owns a computer cluster consisting of 28,936 A100 GPUs.
To scale this up to handle Google's annual search volume, a cluster of 4.1 million GPUs?more than 200 times the current size?would be required. Patel's analysis states that the cost of purchasing GPUs alone would exceed $100 billion (approximately 130 trillion KRW).
WP pointed out, "Companies providing AI models to consumers will have to balance the desire to secure market share with the accumulating financial losses."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
