OpenAI will spend up to $7 billion this year to train and run ChatGPT and new large language models, according to a report by The Information. Of the amount, nearly $4 billion will be used to improve inference – or leasing Microsoft server capacity, which is required to operate ChatGPT.

Also read: OpenAI goes lite, releases cheaper AI model called GPT-4o mini

At least $3 billion is earmarked for training the viral AI chatbot and other models with new data, the report said. The figure includes payments OpenAI will make to publishers to use their content. The company will also need an extra $1.5 billion to cover staff costs for some 1,500 employees.

OpenAI could lose $5 billion in 2024

Inference occurs when live data is run through an AI model, allowing it to solve problems without human input. Per The Information, OpenAI has 350,000 servers powered by Nvidia’s A100 chips for inference. Around 290,000 of these handle inference for ChatGPT and are nearing capacity.

The publication said the estimates are based on previously undisclosed financial statements and interviews with “people involved in the business.” To stay in the artificial intelligence race, OpenAI will need to make more urgent investments, the report says.

In the past, the San Francisco-based firm received a $13 billion investment from Microsoft. It is in a hurry to train new models to match the growing demand for its services. OpenAI rakes in about $2 billion in revenue a year from ChatGPT and could earn $1 billion from charging access to LLMs.

Also read: OpenAI’s SearchGPT set to challenge Google search dominance

However, that may not be enough to meet the growing costs of training its AI models. The Information’s report raises questions about the ability of OpenAI to make a profit soon. Based on current revenue estimates, the company could be left with a $5 billion financial hole, the report says. OpenAI will likely need new funding over the next 12 months, it added.

“Investors should ask: What is their moat?  Unique tech?” AI critic Gary Marcus wrote on X. “What is their route in profitability when Meta is giving away similar tech for free? Do they have a killer app? Will the tech ever be reliable? What is real and what is just demo?” he added.

Anthropic AI, the maker of the Claude chatbot, previously said it could spend as much as $2.7 billion in cash to train its models.