Sacra estimates that Hugging Face hit $70M annual recurring revenue (ARR) at the end of 2023, up 367% from 2022, though a subsequent report placed annualized revenue at $30M in mid-2023 when the company raised at a $4.5B valuation—suggesting the $70M figure reflects a strong second-half acceleration driven by large consulting contracts with Nvidia, Amazon, and Microsoft.

Like Github, Hugging Face makes money from paid individual ($9/month) and team plans ($20/month), but the majority of their revenue has historically come from spinning up closed, managed versions of their product for the enterprise. The company has since begun shifting its revenue mix away from one-time consulting engagements toward more predictable recurring streams, including API usage fees and referral revenue from cloud partners, a transition that included cutting roughly 10 roles from its consulting-heavy Expert Support Program.

Hugging Face reached a valuation of $4.5 billion following funding rounds that have raised a total of $396 million to date. The company has attracted investment from major technology companies including Google, Amazon, Nvidia, IBM, and Salesforce, establishing itself as a significantly funded AI infrastructure company.

Hugging Face (2017) found product-market fit as the Github of ML, allowing any developer with a bit of Python knowledge to quickly get started, with 500K+ public datasets, one-button LoRa fine-tuning bundled into their Transformers library (121K stars on Github), and 2M+ pre-trained open-source models. The platform now serves 13M users across 500,000 organizations, with 30%+ of the Fortune 500 maintaining verified accounts.

Researchers open-source their models and upload them to Hugging Face because while it takes millions of dollars in compute to build a consumer product like ChatGPT, distributing your model via Hugging Face gets it in front of every ML developer for free—and leaves open the potential to charge for an API version of it down the line.

Hugging Face's main product is a library of pre-trained models and tools for NLP tasks. Think of these models as trained experts in understanding and processing language. For example, Hugging Face has models that can classify text into different categories (like sentiment analysis), extract information from text, translate text, generate text, and more.

These models are trained on a large amount of text data, which allows them to learn the patterns and structures of human language. Once the models are trained, they can be fine-tuned on specific NLP tasks with a smaller amount of data in what's known as transfer learning.

Hugging Face also operates a cloud platform that provides a range of NLP and AI services, including model hosting, inference, and optimization, so that users can deploy, manage, and scale their NLP models in a cloud environment.

Hugging Face has expanded its own model releases as a platform signal. SmolLM3 is a 3B-parameter multilingual model trained on 11T tokens with 128k context, dual reasoning modes, tool calling, and six-language support, positioned to outperform comparable open models in its size class. SmolVLA is a 450M-parameter open-source vision-language-action model for robotics that delivers roughly 30% faster task completion and approximately 2x task throughput compared to prior approaches.

Open-source robotics is now a formal product line for Hugging Face, anchored by its acquisition of Pollen Robotics (April 2025). The hardware lineup spans the full market: Reachy 2 is a full humanoid robot at $70,000 already deployed across hundreds of units in 20+ countries, while Reachy Mini brings a consumer-facing robot to market at $299 (Lite) and $449 (wireless), and the SO-101—a 3D-printable robotic arm starting at $100—extends the platform down to hobbyists and researchers. The robotics software stack, LeRobot, has reached v0.5.0—its largest release to date—adding Unitree G1 humanoid support, Pi0-FAST autoregressive VLA policies, real-time chunking, and NVIDIA IsaacLab-Arena integration, with contributions from 50+ new contributors. The platform effect is already visible in the data: robotics datasets on the Hub grew from 1,145 in 2024 to 26,991 in 2025, making robotics the largest dataset category on the platform.

Hugging Face was founded in 2016 as a chatbot for teens, but they found product-market fit in 2018 after they open-sourced the Transformers library they built to run it, consisting of a set of 30 pre-trained models for text classification and summarization with APIs so any developer could quickly train those models on their own data.

On top of that core library, Hugging Face has built a Github-like cloud platform for collaborating on and sharing those models, with 2M+ user-contributed open-source models available for developers to use for training.

Today, Hugging Face's primary sources of revenue include:

Hugging Face's basic product consists of premium access to their database of models, with two basic tiers between "pro" and "enterprise".

Pro costs $9 per month and offers private dataset viewing, inference capabilities, and early access to new features. Pro users also receive $2/month in inference credits usable across Hugging Face's Inference Providers network.

Enterprise is tailored to bigger teams and is priced per-user at $20 per month, and it offers single sign-on (SSO), regional data storage, audit logs, access control, the ability to run inference on your own infrastructure, and priority support.

Hugging Face operates a cloud platform that provides a range of NLP and AI services, including model hosting, inference, and optimization. The platform allows users to easily deploy, manage, and scale their NLP models in a cloud environment. Hugging Face charges its cloud users based on usage, with fees for model hosting, inference, and optimization. The cloud offering includes Inference Providers—a routing and billing layer that integrates third-party inference providers (fal, Replicate, SambaNova, Together AI) directly into Hub model pages and SDKs, billing routed requests through the user's HF account at standard provider rates with no additional markup. Rounding out the cloud stack is Training Cluster as a Service—a joint product with NVIDIA—giving any of Hugging Face's 500,000 organizations on-demand access to large GPU clusters billed only for the duration of a training run.

Hugging Face offers a range of enterprise solutions that leverage its NLP and AI technology. These solutions include custom model training, deployment, and integration services, as well as access to premium features and support. Hugging Face charges its enterprise customers for these services on a contract basis.

Hugging Face has been actively shifting its revenue mix away from one-time consulting contracts—trimming its Expert Support Program headcount—and toward recurring streams: API usage fees, hardware sales from its robotics product line, and referral fees earned when cloud partners such as AWS and Azure deploy Hugging Face models. The Microsoft partnership now puts 10,000+ Hugging Face models into Azure AI Foundry for click-to-deploy use, and Google Cloud usage of Hugging Face has grown 10x over three years, representing tens of petabytes of model downloads per month. Hugging Face also continuously scans its 2.2M+ public repositories via a VirusTotal collaboration, using security compliance as an enterprise differentiator.

Hugging Face's model-agnostic, collaborative workspace provides them with unique positioning in the AI space. 

By allowing teams to e.g. leverage high-quality closed-source models to train their own low-cost, open-source models, Hugging Face has exposure both to the upside of proprietary models getting more advanced and open-source models getting cheaper and faster to train.

While the companies behind individual models may rise and fall, Hugging Face both perpetuates and benefits from increasing non-monogamy among ML developers, who prefer mixing and matching models to find the best solution for their particular use case.

OpenAI hit $2B in annual recurring revenue at the end of 2023, up about 900% from $200M at the end of 2022.

OpenAI was founded in December 2015 as a non-profit dedicated to developing “safe” artificial intelligence. Its founding team included Sam Altman, Elon Musk, Greg Brockman, Jessica Livingston, and others.

OpenAI’s first products were released in 2016—Gym, their reinforcement learning research platform, and Universe, their platform for measuring the intelligence of artificial agents engaged in playing videogames and performing other tasks.

OpenAI’s flagship consumer product today is ChatGPT, which millions of people use everyday for tasks like code generation, research, Q&A, therapy, medical diagnoses, and creative writing.

OpenAI has about two dozen different products across AI-based text, images, and audio generation, including its GPT-3 and GPT-4 APIs, Whisper, DALL-E, and ChatGPT.

Earlier this year, Google merged its DeepMind and Google Brain AI divisions in order to develop a multi-modal AI model to go after OpenAI and compete directly with GPT-4 and ChatGPT. The model is currently expected to be released toward the end of 2023.

Gemini is expected to have the capacity to ingest and output both images and text, giving it the ability to generate more complex end-products than a text-alone interface like ChatGPT.

One advantage of Google’s Gemini is that it can be trained on a massive dataset of consumer data from Google’s various products like Gmail, Google Sheets, and Google Calendar—data that OpenAI cannot access because it is not in the public domain.

Another massive advantage enjoyed by Google here will be their vast access to the most scarce resource in AI development—compute.

No company has Google’s access to compute, and their mastery of this resource means that according to estimates, they will be able to grow their pre-training FLOPs (floating point operations per second) to 5x that of GPT-4 by the end of 2023 and 20x by the end of 2024.

Meta has been a top player in the world of AI for years despite not having the outward reputation of a Google or OpenAI—software developed at Meta like Pytorch, Cicero, Segment Anything and RecD have become standard-issue in the field.

When Meta’s foundation model LLaMA leaked to the public in March, it immediately caused a stir in the AI development community—where previously models trained on so many tokens (1.4T in the case of LLaMa) had been the proprietary property of companies like OpenAI and Google, in this case, the model became “open source” for anyone to use and train themselves.

When it comes to advantages, Meta—similar to Google—has the benefit of compute resources that they can use both for developing their LLMs and for recruiting the best talent. Meta have the 2nd most H100 GPUs in the world, behind Google.

Anthropic is an AI research company started in 2021 by Dario Amodei (former VP of research at OpenAI), Daniela Amodei (former VP of Safety and Policy at OpenAI) and nine other former OpenAI employees, including the lead engineer on GPT-3, Tom Brown. Their early business customers include Notion, DuckDuckGo, and Quora.

Notion uses Anthropic to power Notion AI, which can summarize documents, edit existing writing, and generate first drafts of memos and blog posts.

DuckDuckGo uses Anthropic to provide “Instant Answers”—auto-generated answers to user queries.

Quora uses Anthropic for their Poe chatbot because it is more conversational and better at holding conversation than ChatGPT.

In March 2023, Anthropic launched its first product available to the public—the chatbot Claude, competitive with ChatGPT. Claude’s 100K token context window vs. the roughly 4K context window of ChatGPT makes it potentially useful for many use cases across the enterprise.

There are a variety of routes that Hugging Face can take both to increase its monetization of its existing market and grow that market further.

Similar to early Github, Hugging Face appears to be under-monetized.

Where Github gave away unlimited public repos for free and charged businesses just $25/month to create private repos, Hugging Face lets users host unlimited models, datasets, and spaces, public or private, for free.

The vast majority of their annualized revenue today comes from the managed version of their product they're selling into companies like Amazon, Nvidia, and Microsoft.

The upshot—Hugging Face is betting on there being a much bigger monetization opportunity in the future as a result of being the central collaborative tool for devs building with AI.

Most of what Hugging Face offers today is still aimed at software developers with a working knowledge of Python, though they no longer need to have any real familiarity with machine learning per se to get started.

But Hugging Face can expand their TAM significantly by offering no-code and low-code workflows for spinning up specific kinds of AI-based products for various verticals.

One early example of this is HuggingChat—a web app launched in February 2024 that lets users create personalized AI chatbots for free, leveraging the thousands of text-based language models that have been uploaded to their platform.

Building and launching more of these kinds of apps can allow Hugging Face to go after end-to-end seats within organizations, capturing product managers and others who might be building apps in Zapier and Airtable today.

Hugging Face's acquisition of Pollen Robotics opened an entirely new hardware and data TAM. The product line spans $70,000 industrial humanoids (Reachy 2) down to a $299 consumer robot (Reachy Mini) and a $100 3D-printable arm (SO-101), giving Hugging Face price points across research labs, enterprises, and hobbyists. The deeper opportunity is that physical robots generate proprietary training data: robotics datasets on the Hub grew from 1,145 to 26,991 in a single year, which could entrench Hugging Face as the canonical data and model repository for the emerging embodied AI stack, just as it became the default for language models. LeRobot v0.5.0 (March 2026) added support for the Unitree G1 humanoid and NVIDIA IsaacLab-Arena simulation environments, expanding the addressable hardware ecosystem well beyond Hugging Face's own devices to third-party commercial platforms.