OpenAI’s costs for training AI models and inference tasks in 2024, according to The Information, could reach up to $7 billion. At the same time, the company could record cash losses of $5 billion, which will force it to look for new opportunities to attract investment.
According to people in the know, OpenAI uses the equivalent of approximately 350 thousand servers with NVIDIA A100 accelerators. Of these, about 290 thousand support ChatGPT. It is stated that the equipment is operating at almost full capacity.
Image Source: Pixabay.com
As part of training AI models and inference, OpenAI receives significant discounts from the Microsoft Azure cloud platform. Specifically, Microsoft charges OpenAI about $1.3 per A100 accelerator per hour, which is much lower than usual rates. However, OpenAI could spend about $3 billion in 2024 on training ChatGPT and other models alone.
Today, OpenAI has approximately 1,500 employees and continues to expand its workforce. Payroll and employee costs could reach $1.5 billion in 2024. The company receives about $2 billion a year from ChatGPT and could get another $1 billion or so from charging access to its large language models (LLMs). OpenAI’s total revenue, according to recent results, is $280 million per month. In 2024, the company’s total revenues are estimated to be in the range of $3.5 billion to $4.5 billion.
So, with expected costs of $7 billion for AI training and inference, as well as $1.5 billion in personnel costs, OpenAI could lose up to $5 billion. This is much higher than the projected costs of competitors such as Anthropic (backed by Amazon), which expects costs to be $2.7 billion in 2024. It is possible that OpenAI will try to conduct another round of financing. The company has already completed seven investment rounds, raising a total of more than $11 billion.