GPU calculation stability AI 9 times higher than income
- April 8, 2024
- 0
Stability-AI used GPUs in the cloud at a cost of $99 million per year. By 2023, the company would earn just $11 million in sales. The future of
Stability-AI used GPUs in the cloud at a cost of $99 million per year. By 2023, the company would earn just $11 million in sales. The future of the Stable Diffusion maker remains uncertain.
Mismanagement and lack of funds are damaging Stability-AI and are the direct reason for the departure of founder and ex-CEO Emad Mostaque. This emerges from a detailed analysis by Forbes. Stability-AI is said to have accumulated more than $100 million in debt from cloud providers to train its models, including the new stable Diffusion 3.
Stability, like all other LLM developers, requires huge GPU power to train models. The company follows an open source approach and cannot count on large investors like Microsoft with OpenAI. To do this, it rented GPU infrastructure from AWS, Google and the specialist provider CoreWeave. Forbes reports that the annual bill was $99 million, excluding $54 million in payroll and other costs.
Debts
However, there were no new financial injections and the commercialization of Stable Diffusion did not go particularly well. For commercial customers, Stability-AI charged $20 per month, but revenue only generated $11 million (estimated) in 2023. That’s not bad, but it’s nine times less than the annual cloud bill alone. The company is said to have accumulated large amounts of outstanding debts from all providers.
In addition, CEO Mostaque was generously endowed with GPU computing power. For example, 2.5 million GPU computing time went into external projects that ultimately produced nothing. Various plans to reduce debt had little success. Mostaque was finally forced to resign last month.
Uncertain future
Stability-AI is now under new management, but it is unclear whether and how all debts have already been paid. In addition, the AI company is facing lawsuits, including from Getty for violating copyright in stable diffusion training.
The episode shows how expensive LLM training is. Debts quickly rise into the millions, making it very difficult to create large LLMs without the support of leading technology partners with deep financial resources.
Source: IT Daily
As an experienced journalist and author, Mary has been reporting on the latest news and trends for over 5 years. With a passion for uncovering the stories behind the headlines, Mary has earned a reputation as a trusted voice in the world of journalism. Her writing style is insightful, engaging and thought-provoking, as she takes a deep dive into the most pressing issues of our time.
