Executive Summary:
Hugging Face develops artificial intelligence algorithms and products aimed at software developers in the machine learning space.
Hugging Face makes money from different licensing fees as well as by selling branded merchandise via its own online store.
Founded in 2016, Hugging Face is one of the most popular communities in the field of machine learning. Its business is currently valued at $2 billion.
What Is Hugging Face?
Hugging Face develops artificial intelligence (AI) algorithms and products aimed at software developers.
It essentially acts as a community and platform for anyone interested in data science. The majority of the models it makes available on an open-source basis are situated in the field of natural language processing (NLP).
Furthermore, Hugging Face offers thousands of data sets, such as IMDb movie scores, that developers can use to train their models on.
Meanwhile, developers can exchange tips and tricks with each other or share their machine learning applications via Hugging Face’s dedicated spaces.
All the work that users work on is stored within a dedicated GitHub repository, which can be freely accessed by anyone in the community (if the developer chooses to open it up).
Lastly, Hugging Face itself also sells enterprise-level products to business customers, including models that don’t require code (AutoTrain) or Private Hub, which enables organizations to collaborate in private.
Detailing the Founding Story of Hugging Face
Hugging Face, headquartered in New York City, was founded in 2016 by Clement Delangue, Julien Chaumond, and Thomas Wolf.
Delangue, Hugging Face’s CEO, grew up in a small town north of France where there wasn’t much to do besides being on his computer, which he was gifted at the age of 12.
By the time he turned 17, Delangue was one of the biggest French merchants on eBay after he began selling ATVs and dirt bikes sourced from China.
He became so successful that eBay even offered him an internship when he started university in Paris. eBay also became the place where he first learned about the prowess of AI.
The online marketplace sent him to an e-commerce trade show where he wound up meeting the cofounder of Moodstocks, an image-recognition startup that used machine learning to scan barcodes and other visuals.
“With a very small team, they were managing to do stuff on par with what Google was doing with 100 times more people,” he recalled about the startup in an interview with Forbes.
Delangue ultimately declined his eBay job offer and joined Moodstocks, which was acquired by Google in 2012. After dabbling in a startup idea of his, he ultimately decided to join mention, another startup, as its Chief Marketing Officer.
After mention was acquired as well, Delangue finally felt ready to give this entrepreneurship thing another go. He then recruited Julien Chaumond, another fellow entrepreneur, and Thomas Wolf, a college friend of Chaumond who earned his Ph.D. in physics, to complete the trifecta.
At the time, chatbots were all the rave, leading to various companies receiving funding to build software for both consumers and businesses.
So, after months of refinement and armed with $1.2 million in angel funding from Betaworks, the trio released an iOS chatbot app aimed at teenagers in March 2017.
“We’re building an AI so that you’re having fun talking with it. When you’re chatting with it, you’re going to laugh and smile — it’s going to be entertaining,” said Delangue when the app launched.
Over the coming two years, the team continued to refine the app’s user experience while raising another $4 million in seed funding (05/2018). And while users eventually began to exchange over 1 million messages a day with their bot friend, monetizing it was an entirely different story.
However, a substantially more potent technology would prove to be the opportunity that the founding team capitalized on.
Back in June 2017, scientists at Google Brain released a paper called Attention Is All You Need where they introduced Transformers, a novel type of neural architecture vastly superior in almost every regard.
Not only did Google use the tech to vastly improve its search results but startups like OpenAI rode it all the way to a $29 billion valuation.
Naturally, Hugging Face hopped onto the trend as well. In 2018 and 2018, it released various transformer libraries for frameworks such as PyTorch, TensorFlow, and BERT. Over 1,000 companies began using the PyTorch library just months after it was released.
VCs tried to capitalize on Hugging Face’s hype and poured another $15 million into the firm in December 2019, valuing it at $80 million. The startup’s various frameworks quickly rose to one of the most-rated GitHub repositories and a staple among machine learning enthusiasts.
By early 2021, Hugging Face was well on its way to becoming the de-facto community for NLP developers. When it raised $40 million in Series B funding in March of that year, its various GitHub repositories had already been forked over 10,000 times.
At the time, Hugging Face was already profitable and thus did not need the money. Over the coming months, Hugging Face continued to expand the reach of its platform, either by partnering with cloud providers like Amazon’s AWS or by releasing hundreds of unique datasets for training purposes.
Its exponential growth was rewarded with yet another round of funding. This time, investors poured $100 million into Hugging Face while valuing it at an eye-popping $2 billion.
Meanwhile, the founding team also turned down multiple “meaningful acquisition offers” and instead doubled down on expanding Hugging Face’s reach, for example by partnering with Microsoft and directly integrating its models into Azure.
Today, Hugging Face employs close to 200 people across offices in New York City and Paris while its models are being used by thousands of companies.
How Does Hugging Face Make Money?
Hugging Face makes money from different licensing fees as well as by selling branded merchandise via its own online store.
Let’s explore each of the firm’s revenue streams in the section below.
Licensing Fees
The overwhelming majority of revenue that Hugging Face generates comes from the various licensing fees it charges for using its models and software.
Hugging Face’s pricing can be accessed here. For example, Inference Endpoints, which allow developers to deploy any ML model on dedicated and autoscaling infrastructure, costs $0.06 per hour.
Meanwhile, a so-called Pro account, which comes with a dedicated badge, early access to new features, and higher tiers for AutoTrain and other features, costs $9 per month per user.
However, most of the money is being made with Hugging Face’s Enterprise solution, with pricing being dependent on the individual use case.
Enterprise customers receive guidance from the firm’s ML experts (such as co-founder Thomas Wolf), get access to a dedicated and secure private deployment hub, and more.
Hundreds of companies, ranging from small startups like Grammarly to tech giants like Google or Salesforce, are customers of its enterprise solution.
Merchandise Sales
Hugging Face also sells branded merchandise, ranging from hoodies to baseball caps, via its own branded online store.
While the selling of merchandise will only represent a small chunk of the firm’s overall revenue, it still is fairly profitable in all likelihood (given the prices those products are sold).
More importantly, selling merchandise will act as another marketing channel and allows fans of a given community to express their admiration in other ways.
It certainly isn’t uncommon for tech companies to sell branded merchandise. Others, such as SpaceX, have even been deriving substantial revenue from selling items online.
The Hugging Face Business Model Explained
The business model strategy that Hugging Face pursues is based on becoming a platform for the machine learning community.
“They saw that transformer-based models working their way outside of NLP and saw a chance to be the GitHub not just for NLP, but for every domain of machine learning,” said Sequoia partner Pat Grady about the firm’s strategy.
Indeed, the adoption of transformers technology has become so far-reaching that it is now applied across a variety of different machine learning domains, whether that’s NLP, image recognition, and even robotics.
Much like GitHub, Hugging Face prioritized (and still does) adoption over monetization. By being the default community of choice for machine learning developers, it then has a much easier time selling into the enterprise, which has become its bread and butter.
Today, over 5,000 organizations are already using Hugging Face. Microsoft, for example, has released over 230 models fine-tuned models alone.
And the more those firms rely on Hugging Face to advance their NLP development, the unlikelier they are to churn.
Hugging Face then boosts adoption via a variety of different growth hacks. The previously-mentioned data sets inspire developers to explore ML-based models. The startup has also released a free course on Reinforcement Learning, which aims to expand the availability of new developers.
Meanwhile, it also uses so-called tasks, which bring together developers to solve complex problems in computer vision, NLP, audio, and more. Hugging Face could theoretically also monetize those challenges by having companies sponsor the task (similar to what Kaggle does).
Another way in which Hugging Face expands its platform is through partnerships. Its various models are made available on AWS and Azure, which in turn also strengthens the quality of those cloud platforms.
Microsoft is certainly no stranger to partnering with AI-based companies. It is now deeply integrated with ChatGPT-maker OpenAI. In return, OpenAI gets to access Microsoft’s compute at discounted rates in an effort to advance its own platform-based business.
Hugging Face also allows developers to set up their own profiles where, similar to GitHub, they can display the models or datasets they’ve deployed. This acts as a badge of honor, thus upping their employability and, in turn, incentivizing users to publish even more content.
And the more content (models, datasets, etc.) is being published on Hugging Face, the more attractive it becomes for enterprise customers because they can access that open-source work to advance their own products.
Hugging Face Funding, Revenue & Valuation
Hugging Face, according to Crunchbase, has raised a total of $160.2 million across five rounds of venture funding.
Notable investors include Lux Capital, Addition, Betaworks, Sequoia Capital, and wealthy angels like Kevin Durant.
Hugging Face is currently valued at $2 billion (post-money) after raising $100 million in Series C funding back in May 2022.
And although Hugging Face does not disclose revenue figures to the public, Forbes was able to verify that the firm generated $10 million in income throughout 2021.