Stability AI faced financial difficulties due to huge cloud infrastructure costs

Mathieu Stern/Unsplash

UK-based generative model developer Stable Diffusion's generative model company Stability AI has found itself on the verge of bankruptcy due to the out-of-control costs of renting the cloud computing power needed to train its AI models.

Here's What We Know

According to an investigation based on internal company documents, by October 2023, Stability AI was left with just $4 million in reserves after spending about $99 million a year just to rent infrastructure from Amazon Web Services, Google Cloud, and other providers.

Meanwhile, much of the leased computing resources were provided to third-party researchers looking to experiment with Stability AI models. One of the company's projects spent at least $2.5 million over four months.

The huge infrastructure costs were not matched by adequate revenues. Stability AI is projected to generate just $11 million in revenue in 2024, while operating expenses exceed $50 million.

The company owed cloud providers at least $9.6m in unpaid invoices for July and August 2023, and was considering deferring tax payments in the UK rather than cutting employee salaries in the US.

The funding shortfall has undermined investor confidence. Former CEO Emad Mostaque had hoped to raise $95m but was only able to raise $20m from Intel against a promised $50m.

After a series of fundraising failures, Stability AI switched to a paid subscription for commercial users of Stable Diffusion in December. The company also considered reselling its cloud resources as a revenue-generating business.

Mostak announced his resignation at the end of March, recognising the problem of "concentration of power in AI". Stability AI is now led by acting CEOs, and its future remains uncertain amid financial difficulties and copyright infringement lawsuits.

Source: The Register