The Hugging Face website on a smartphone arranged in New York, Aug. 17, 2023.
Gabby Jones | Bloomberg | Getty Images
Hugging Face, an AI firm based in New York, has raised $235 million at a $4.5 billion valuation from some of technology’s biggest companies.
Google, Amazon, Nvidia, Salesforce, AMD, Intel, IBM and Qualcomm all contributed to the round, the company said. Hugging Face CEO Clement Delangue said the funds are to be focused on hiring talent to be competitive in the artificial intelligence space.
Startups working on AI models have reached high valuations as big companies and venture capitalists seek to plow money in the recent AI boom, which kicked off last year when Microsoft-backed OpenAI released its ChatGPT chatbot.
Hugging Face’s big valuation and crop of prominent backers reflect how a more collaborative approach to building AI has been gaining steam in recent months, especially after Facebook parent Meta released its Llama large language model, which is free to use for the vast majority of companies.
Other highly valued AI startups, like OpenAI or Cohere, work on the technology directly and guard the results as a trade secret, then charge customers to access them through application programming interfaces, or APIs.
But Hugging Face produces a platform where AI developers can share code, models, data sets, and use the company’s developer tools to get open-source artificial intelligence models running more easily. In particular, Hugging Face often hosts weights, or large files with lists of numbers, which are the heart of most modern AI models.
While Hugging Face has developed some models, like BLOOM, its primary product is its website platform, where users can upload models and their weights. It also develops a series of software tools called libraries that allow users to get models working quickly, to clean up large datasets, or to evaluate their performance. It also hosts some AI models in a web interface so end users can experiment with them.
It’s similar in theme and practice to code-repository GitHub (which Microsoft acquired in 2018), where coders from around the world post their projects while they’re working on them.
Hugging Face endorses the belief that most companies working with AI will want to develop their own models or technology, and will need tools to do so, co-founder and CEO Delangue told CNBC. He hopes that AI developers will rely on Hugging Face on a daily basis to get their work done.
One reason the big companies are investing: Their employees are actively using the platform, he said.
“AI builders are using Hugging Face all day, every day,” Delangue said. He predicted that the number of software developers working with AI models would grow in the coming years.
“Maybe in five years, you’re going to have like 100 million AI builders. And if all of them use Hugging Face all day, every day, we’ll obviously be in a good position,” he said.
Although most attention in recent weeks has been on so-called large language models like ChatGPT or Llama that focus on generating text, Hugging Face hosts any AI model, including ones that generate music or images, translate languages, or identify objects inside images. Hugging Face hosts 500,000 different AI models, 250,000 data sets, and has 10,000 paying customers, the company said.
Delangue told CNBC he had recently played with IDEFICS, which allows users to upload an image and ask questions about it. He also mentioned Seamless M4T, a translation model from Facebook, as well as Llama2, Facebook’s language model.
What’s with the funny name?
Hugging Face is named after an emoji, the hugging face, a smiley face framed by two open hands.
The name and logo date back to the company’s founding. Hugging Face was originally a iPhone chatbot app, but when the company open-sourced some its machine-learning code, it realized that it was catching on with AI developers, and pivoted toward that.
“When we started the company, with my cofounders Julien Chaumond and Thomas Wolf, we joked that we wanted to be the first company to go public with an emoji instead of the three letter ticker,” Delangue said.
“Maybe during this round we should start our lobbying exercise with the Nadsaq for them to allow us to use emojis on their board,” he quipped.
