London-based startup Stability AI has raised less direct funding than has previously been reported, as a “majority” of its last funding round came in the form of a “compute for equity” style deal, Sifted has learned.
The investment from Intel — reported as being nearly $50m by Bloomberg in November last year — saw Stability committing to using a majority of the round to purchase access to the US chip company’s hardware resources, according to three people with knowledge of the deal. A minority part of the investment round, which included other smaller investors, did not involve hardware purchase commitments, the sources said.
Stability will get access to Intel’s compute resources via a new, AI-optimised supercomputer that the two companies have announced they are co-developing.
Compute for equity deals have become a common feature of the generative AI funding landscape, with both OpenAI and Anthropic raising money on similar terms from Microsoft and Amazon respectively in the last year.
These deals are a useful way for AI companies to avoid paying for expensive hardware from their balance sheets — but don’t help cover some of the other sizeable costs AI startups have, like talent.
Excluding this latest reported funding round from Intel, Stability AI has raised $101m since October 2022 and employs more than 190 people, according to Dealroom data. That’s far less funding raised, and far more people on the team, than some of its competitors. France’s Mistral, by comparison, employs 51 people and has raised $537m since June 2023, while Germany’s Aleph Alpha employs 84 people and has raised $641m since launching in 2019.
To cover costs, Stability is trying to open up new revenue streams, and is reportedly looking for a buyer to shore up its ability to keep developing costly generative AI models.
Intel declined to comment to Bloomberg on the deal in November, and didn’t respond to Sifted’s request for comment for this story. Stability also declined to comment.
Stability x Intel
In December 2023 Stability detailed its plans to build an AI-optimised supercomputer with Intel. It said that it wanted to “explore alternatives to GPUs” (the type of chip NVIDIA specialises in producing that has become industry standard for AI training companies), and that Intel’s new Gaudi 2 AI accelerator chip offered an “economical price point” compared to “its contemporaries”.
Intel has recently described the Stability as its “anchor customer” for its new AI supercomputer.
One former Intel employee, describing the majority part of the investment that came with hardware purchase commitments, said that “Intel practically spends nothing because they were going to build that supercomputer anyway, and they wanted someone to use it.”
“Stability can’t pay salaries with that part of the investment,” they added.
The need for cash
In November Bloomberg also reported that Stability was “exploring a sale”, following “increased pressure from investors over its financial position”. A company spokesperson told the publication that “we are not trying to sell the company and are focused on releasing leading models.”
In December the startup launched a subscription fee, pivoting away from its open source business model. Stability told Sifted that revenue from this new membership model would allow the company “to drive even greater efficiency and deliver better and more impactful foundation models for businesses and researchers worldwide.”