The story of Replicate, a $350M company powering million-dollar AI businesses

First, meet Ben Firshman.

Ben (great name) is the founder and CEO of Replicate. They just raised $40M Series B from a16z and Sequoia at a $350M valuation.

So what’s Ben’s story?

Here is a timeline:

• 2008 - 2011: Firshman studied Computer Science at the University of Warwick

• Jun 2008 - Aug 2008: Summer of Code student at Google where he built a search API for Django

• Dec 2009 - Jan 2010: Developer at Guardian Group where he built the Guardian’s World Government data site.

• Mar 2010 - Oct 2010: Editor in Chief of The Boar (Warwick’s campus newspaper)

• Aug 2010 - May 2011: Developer at Guardian Group where he invented and helped build a new type of publishing system for The Guardian's iPad (and won awards)

• Jan 2011 - March 2012: Firshman co-founded Epio, ‘Heroku for Python’

• Aug 2011 - March 2012: Developer at Lanyrd

• Mar 2012 - May 2012: Firshman joined This Is My Jam as a Developer and built their Spotify app.

• May 2012 - May 2013: He co-founded Poetica Ltd, a writing feedback tool

• Jul 2013 - Jan 2014: Developer at Government Digital Service

• May 2013 - Jun 2014: He founded Orchard, an EC2 for containers, and created Fig, which later was acquired by Docker and evolved into Docker Compose

• Jun 2014 - Dec 2016: Firshman joined Docker, Inc as a Product Manager and later became the Director of Product Management, leading the development and management of Docker's open-source projects

• Jan 2017 - Dec 2018: He took a break from the tech industry to travel the world, cycling across Europe and India, and driving from London to China

  • while on the road he built arXiv Vanity which turns academic papers into web pages

• Oct 2019 - Present: Firshman founded Replicate

pfft…overnight success 👀

Some fun things I found about Ben:

Starting from the beginning of Replicate

Looking at his GitHub history, he always contributed publicly to repositories over the years. Still, between October 9th and Jan 28th 2019 (when he ‘joined’ the Replicate organisation), there were 277 contributions to private repos.

They raised a pre-seed on Jan 1st 2020 from Y Combinator.

The first commit to a public Replicate repo was on Jul 1st 2020 and subsequently, his cofounder, Andreas started committing on the 3rd.

This was for ‘Keepsake’, which started the Replicate journey.

This site looks the same today as it did back then.

Work on the Replicate website started on the 6th July 2020.

lt was built on Docusaurus and hosted on replicate.ai - the first snapshot of the site I could find was from the 4th of November 2020. Look familiar?

They started with two examples:

• getting started with replicate - the model used in the getting started guide

• movie recommendations - (you can guess what this did, right?!)

By June 2020 they’d tweaked their offering to run and re-train machine learning models.

Also, a fun snapshot to see how users don’t always get what you do or how to do it… Ben’s commits on making it clear and blindingly clear 😂

Still getting ever clearer about ‘what they do’ 😂

There are so many instances of tiny tweaks to the wording on the homepage to get it right. We all do this, just know we’re not all crazy for it…

Keep explaining it better… 😂

Remember, this was 3 years ago and they were shipping commits on GitHub almost every day.

Finally, Ben tweeted about it on 19th November 2020:

They turned from version control into hosting models in early/mid-2021.

By November 2021, they had other models you could run:

And a new teammate, Zeke joined end of 2021.

Now skip forward a year, the website looks mostly the same - I bet that saved Ben’s sanity. And the new domain replicate.com was in their possession and it was live by November 2021.

The site stayed fairly consistent for another year, and until 2023 when it changed again, but no tweaks to ‘what it does’.

They raised a Series A of $17.8 million from a16z, Y Combinator, Sequoia and angels on Feb 21st 2023.

They raised a Series B of $40 million announced on December 5th 2023, led by a16z.

Ok, enough about the timeline of Replicate.

They help developers run an AI model, and train them. If you want to make an AI avatar app, you can use one of the models available on Replicate to generate the image for you.

When you get to a certain point, you want to train your own model to tweak the outputs, so you can use Replicate to do that too.

They have over 2 million developers (30k of which are paying customers) and over 25k AI models you can use. They started with images but now have everything including video, audio, and text.

How do they make money?

When you use a model on Replicate, you pay for the time it’s processing your request.

A lot of developers use Replicate for their side projects but side-projects can turn into multi-million dollar companies. Like Pieter Levels and Danny Postma.

• Pieter Levels uses Replicate for his PhotoAI.com site. And makes several million per year.

• Danny Postma uses Replicate for his Headshotpro.com site. And makes over $1M per year.

Replicate has also seen people build autonomous robots, real-time drawing apps, and language model command-line interfaces.

Even small projects within big companies can grow to become real products.

Unsplash is labelling all the images in their catalog using BLIP. BuzzFeed is… turning your pets into plushies. Character AI, Labelbox, and thousands of other businesses are deploying models on Replicate.

(source: replicate blog)

They’re going to spend that $40M on:

• Making it better for businesses

• Making it faster

• Better language models (I guess a big update may be their own?)

The takeaway

Side-projects can turn into million-dollar businesses. Replicate certainly has.

There are many hours of coding, commits, website changes, making things clearer, edging in slightly different directions etc behind what looks like an easy journey to raising a ton of money and being worth $350M. Okay, not easy easy but ya know…

Sometimes the thing you’re working on doesn’t feel like it could be big. But sometimes it can be.

To build the final thing you actually have to start building a thing.

You start realising when you’re building what needs to be built. Maybe that’s from user feedback, your gut, no one using it at all or something else.

Keep building stuff.

And if you don’t know how. Build a no-code website with Carrd. Automate a process with Zapier. Sell a digital PDF on Gumroad. Write a newsletter about AI anything.

Start with whatever you can.

And if you like the design of a website you made, you can just keep using it. For a few years at least… until you have 30k customers you can make something slick like this…

Please let me know what you think of this kind of post and if I should do more.