Leaked deck raises questions over Stability AI's pitch

When generative AI darling Stability AI closed a massive $101 million Series A round in October 2022, Silicon Valley was stunned. CEO Emad Mostaque, a former hedge fund manager, was virtually unknown in deep learning circles. Yet his company seemed to control Stable Diffusion, the open image model taking on OpenAI.

Now, a leaked investor deck from July 2022 obtained by Sifted reveals how Mostaque convinced deep-pocketed venture capitalists to buy in: by taking credit for other people's academic work.

Packaging open research as proprietary tech

According to the pitch deck, Stability AI claimed to be developing proprietary generative models, pitching Stable Diffusion as its own in-house breakthrough.

The reality was far different. The foundational research behind Stable Diffusion was published in April 2022 by researchers at LMU Munich and Runway. Stability only entered the frame later, largely paying for compute through Amazon Web Services credits rather than authoring the architecture.

Stability also claimed to "co-develop" leading open-source models with external groups like EleutherAI. In pitch conversations, Mostaque went as far as claiming Stability "owned" EleutherAI and teasing massive unreleased models. When asked if Stability owned the collective, a core EleutherAI member gave Sifted a blunt one-word response: "Nuts."

The crumbling supercomputer moat

Mostaque's most potent selling point was brute-force infrastructure. The deck boasted Stability possessed the "world’s 7th largest supercomputer," packing over 4,000 Nvidia A100 GPUs and out-computing rivals by a factor of five.

That claim ignored a massive catch. Stability had simply accessed "spot instances" via AWS—discounted compute capacity that Amazon can revoke at a moment's notice whenever demand spikes.

Björn Ommer, the LMU professor whose lab pioneered the model, noted the ethical dilemma. He pointed out that raising massive rounds by telling investors you built a system that largely came out of academia creates severe scientific and credibility issues.

Stability pushed back, telling Sifted it was an organizer in the early development phase and never hid academic credits.

Why it matters

Stability AI pulled off one of the fastest venture scale-ups in AI history, but this leaked deck shows it was built on exaggerated foundations.

For enterprise users and developers, this is an immediate amber alert. If an infrastructure provider’s core tech is borrowed and its supercomputing moat is just rented AWS capacity, relying on them for mission-critical operations looks increasingly hazardous.

For the AI ecosystem, it exposes the toxic side of the generative AI hype cycle. Stability acted as a slick broker: fund an open-source academic team, rebrand their milestone as an in-house asset, and spin compute arbitrage into a billion-dollar valuation.

In our view, Mostaque played venture dynamics brilliantly in the short term, but the blowback was inevitable. Alienating academic collaborators like LMU and burning bridges with grassroots groups like EleutherAI turns an open-source messiah into a pariah. True moats require fundamental research, not just creative slide decks and cloud credits.