• Today in AI
  • Posts
  • Stability: The AI unicorn betting on decentralized AI

Stability: The AI unicorn betting on decentralized AI

This AI unicorn is betting on decentralized AI.

They’ve raised over $170M at $1B+ valuations.

Their main product has over 300M downloads.

So what’s the startup and what do they do?

Stability AI is an AI startup focused on creating the best quality LLM models that you can take, own and adapt to your own needs.

This makes it one of the only open-source focused AI startups along with Mistral AI.

Their mission is to “activate humanity’s potential through generative AI” and have these technologies accessible to everyone everywhere.

Stability AI was founded in 2020 by Emad Mostaque, and started as a nonprofit working on projects like COVID research, doing a lot of pro bono work.

Here’s a bit of background on Emad:

He was born in 1983 to a Bangladeshi muslim family.

Supposedly holds a BA and MA in math and CS from Oxford University. However he allegedly never showed up to attend his graduation ceremonies so hasn’t officially received his degrees.

He is diagnosed with Asperger’s and ADHD (pretty strong sign for a genius).

In Emad’s twenties he created online forums for muslim communities to help the Islamic world. He also created “Islamic AI” which helps guide people on their islamic journeys.

He started his career as a hedge fund manager in crude oil trading and even gave advice to governments in the middle east on affairs and Islamic Extremism.

In 2019, Emad founded Symmitree - a startup that reduced the cost of technology for people living in poverty. He worked on this for a year before starting Stability.

From this information alone, Emad has a pretty impressive track record of making the world a better place.

This makes him the perfect advocate for leading open-source AI, something that could be extremely dangerous.

So, what’s the business?

The startup was initially self-funded by Emad who had a vision of making generative AI accessible to everyone.

Stability later introduced “Stable Diffusion” a gen AI application that turns text prompts to images. Stable Diffusion allows you to type in a phrase and instantly create AI generated images. 

The image generator was originally created by a team from the German university LMU. Development of the tech behind Stable diffusion was led by Patrick Esser of Runway and Robin Rombach of CompVis.

Stable Diffusion was Stability AI’s biggest product by far and provided a catalyst for rapid growth. The product has recently reached 330 million downloads, a milestone beating Google’s open model downloads.

According to huggingface - the Github for AI devs, Stability was the most downloaded and liked model by developers in November 2023. It was downloaded over 49.8M times.

Stability model downloads even surpassed OpenAI models'…

The unicorn focuses on “Amplified Human Intelligence” rather than AGI which the majority of AI research labs like OpenAI use. Focusing on “AHI” allows people to tell better stories through prioritizing distribution and being open to optimizing models.

But does Stability make money?

Here’s the business model:

  • DreamStudio APIs: Stability AI monetizes Stable Diffusion through DreamStudio’s APIs. This allows users to access Stable Diffusion via an API platform. This is priced based on API usage and the amount of compute required to fulfill a request.

  • Consulting Services: Stability AI’s team uses its expertise in AI to provide consulting services, helping businesses implement AI into their systems.

  • Enterprise Services: The company offers customized services to enterprise/business clients, adapting Stable Diffusion or other models to meet specific needs.

Stability also has partners like Amazon’s AWS. 

AWS is Stability’s preferred cloud provider for building and scaling its models, which probably includes favorable terms to reduce infrastructure costs.

According to Emad, Stability was on track to making $3M in revenue in November 2023 alone. This is however only a fraction of the $8M on bills and payroll costs - Stability is burning through cash.

According to LinkedIn, Stability employs over 200 people. In comparison, competitors like Mistral AI only employ closer to 30 employees.

Stability Diffusion’s biggest competitor is probably Midjourney, another AI image generator that does $300M+ in revenue with only 40 employees.

The startup has raised a total of $176M+ to keep up with burning all this cash. They've reportedly been trying to raise more funding at a $4B valuation since March.

Decentralised AI

The founder Emad recently left Stability AI, stepping down from CEO and the board. 

According to Emad he stepped down to focus on building decentralized AI - he doesn’t like being a CEO and hates dealing with CEO responsibilities like operations, HR, and business development.

Emad is a better strategist and better at designing systems.

Here is why building decentralised AI is so important:

  1. Availability and Accessibility - Gen AI applications should be open and accessible to everyone, everywhere

  2. Governance of data - We should be aware of who manages and controls the data fed to Gen AI systems and need more transparency

  3. How it all comes together - Is it a single package or is it a modularized infrastructure that anyone can build on? Does it require servers? Etc.

The major problem behind the globalization of AI models is that once a government embraces centralized technology, it’s super difficult to decentralize it.

Every country needs an AI strategy. This is what Emad is working on.

Emad has also joined Render Network, a network that connects artists looking to render 3-D media with GPUs in a decentralized network.

Render is allocating over $250M to decentralised data creation and augmentation to enable Decentralized AI.