OpenAI may lose a massive $5bn this year, need more funding?

According to tech media outlet The Information, citing undisclosed internal financial data and insights from relevant business personnel, OpenAI may incur a staggering loss of $5 billion this year. If their analysis holds true, the company may require additional funding within the next 12 months.

This begs the question: if AI companies consistently struggle to reduce the costs of developing and operating AI, do they have no choice but to pass on higher prices to consumers? However, the industry remains confident that the costs of inference will continue to plummet significantly.

So, where exactly is OpenAI spending its money?

Server Costs: The Information cites insiders stating that as of March this year, OpenAI anticipates spending nearly $4 billion on renting Microsoft servers to support ChatGPT and its underlying Large Language Models (LLMs), also known as inference costs.
Training Costs: Beyond running ChatGPT, training costs, including data expenses, are projected to soar to 3 billion this year. Last year, OpenAI accelerated the training of new AI models beyond initial plans, resulting in costs far exceeding the company′s 800 million forecast. This year, as OpenAI trains new versions of its flagship LLMs and initiates training for new flagship models, related training costs could double.
Salary Costs: With a workforce of approximately 1,500 employees, OpenAI spends roughly $1.5 billion on salaries as it competes with giants like Google for top talent. With around 200 job openings listed on its website, the company may hire more employees in the second half of the year.

ChatGPT Revenue: ChatGPT generates approximately $2 billion in annual revenue.
Developer Fees: OpenAI also charges developers, generating over $80 million in monthly revenue as of March.
Total Revenue: Recently, the company's monthly revenue has reached approximately 283 million, translating to anannual revenue range of 3.5 billion to $4.5 billion, contingent upon sales performance in the second half.
Net Loss Projection: Deducting approximately 8.5 billionin operational costs from the higher end of the revenue estimate (4.5 billion) yields a potential loss of 4 billion to 5 billion.

Despite this significant revenue-expenditure gap, the industry sentiment remains relatively sanguine. Ion Stoica, co-founder of enterprise software company Databricks, expressed confidence that the costs of inference will continue to decline significantly, suggesting that OpenAI may reap reasonable profits from its models in the coming years. As for training costs, which are more complex, companies are likely to continue investing heavily to train larger models.